<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Samar Shetye</title>
    <description>The latest articles on DEV Community by Samar Shetye (@samar_shetye).</description>
    <link>https://dev.to/samar_shetye</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3755023%2F12f019fd-d194-4291-905e-f08c7a96a3dc.jpg</url>
      <title>DEV Community: Samar Shetye</title>
      <link>https://dev.to/samar_shetye</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/samar_shetye"/>
    <language>en</language>
    <item>
      <title>RepoSeance: Resurrecting Dead Code with Local Open-Weight AI</title>
      <dc:creator>Samar Shetye</dc:creator>
      <pubDate>Sat, 31 Oct 2026 12:19:18 +0000</pubDate>
      <link>https://dev.to/samar_shetye/reposeance-resurrecting-dead-code-with-local-open-weight-ai-42kc</link>
      <guid>https://dev.to/samar_shetye/reposeance-resurrecting-dead-code-with-local-open-weight-ai-42kc</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-week4-2026-10-26"&gt;Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What I Built
&lt;/h2&gt;

&lt;p&gt;Every software engineer inherits a codebase that feels haunted.&lt;/p&gt;

&lt;p&gt;You clone an archived repository or checkout an unfamiliar branch, and you are immediately greeted by the restless spirits of past developers: cryptic // TODO: fix this before Friday comments written five years ago, half-implemented refactors abandoned mid-stride, and git commit messages like "wip: it works on my machine".&lt;/p&gt;

&lt;p&gt;Usually, the institutional knowledge died when the original authors left the company or archived the repository.&lt;/p&gt;

&lt;p&gt;RepoSeance is an open-source, local-first software archaeology tool. Rather than treating legacy or abandoned repositories as landfill, RepoSeance conducts a digital séance over your git history. It reconstructs project chronology, indexes unmerged branch inflection points, scans for orphaned TODO/FIXME markers, and summons an interactive, evidence-grounded AI assistant powered completely offline by Google's open-weight Gemma 2 model.&lt;/p&gt;

&lt;p&gt;Ask RepoSeance why a module was rewritten, what features were abandoned, or where the riskiest architectural debt lies—and it will answer you with verified citations linking to exact commit SHAs, line numbers, and author timestamps.&lt;/p&gt;




&lt;h2&gt;
  
  
  Demo
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Youtube Demo Video:
&lt;/h3&gt;

&lt;p&gt;  &lt;iframe src="https://www.youtube.com/embed/K178dm6TmvY" width="710" height="399"&gt;
  &lt;/iframe&gt;
&lt;/p&gt;

&lt;h3&gt;
  
  
  Screenshots Walkthrough
&lt;/h3&gt;

&lt;h3&gt;
  
  
  1. Minimalist Landing Page
&lt;/h3&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fwgmeg5z8htnr5vnf04tg.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fwgmeg5z8htnr5vnf04tg.png" alt=" " width="800" height="425"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  2. Repository Overview &amp;amp; Health Telemetry
&lt;/h3&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ffd7cqatjrs9d4kkp9j7k.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ffd7cqatjrs9d4kkp9j7k.png" alt=" " width="800" height="425"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  3. Grounded Chat &amp;amp; Side Evidence Drawer
&lt;/h3&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ffwvprcpsuv25v449734a.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ffwvprcpsuv25v449734a.png" alt=" " width="800" height="425"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  4. Ghost Timeline &amp;amp; Milestone Inflection Points
&lt;/h3&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F8adcmj6ujgr9a9wwahtu.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F8adcmj6ujgr9a9wwahtu.png" alt=" " width="800" height="425"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  5. Issues &amp;amp; Marker Archaeology
&lt;/h3&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fflr6ler5xqejwxkccgb3.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fflr6ler5xqejwxkccgb3.png" alt=" " width="800" height="425"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  6. Automated Resurrection Report &amp;amp; Feasibility Codex
&lt;/h3&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fyjluk3s09q3zsp7bigd4.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fyjluk3s09q3zsp7bigd4.png" alt=" " width="800" height="425"&gt;&lt;/a&gt;&lt;/p&gt;




&lt;h2&gt;
  
  
  Code
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Github Repository
&lt;/h3&gt;


&lt;div class="ltag-github-readme-tag"&gt;
  &lt;div class="readme-overview"&gt;
    &lt;h2&gt;
      &lt;img src="https://assets.dev.to/assets/github-logo-5a155e1f9a670af7944dd5e12375bc76ed542ea80224905ecaf878b9157cdefc.svg" alt="GitHub logo"&gt;
      &lt;a href="https://github.com/Samar-365" rel="noopener noreferrer"&gt;
        Samar-365
      &lt;/a&gt; / &lt;a href="https://github.com/Samar-365/RepoSeance" rel="noopener noreferrer"&gt;
        RepoSeance
      &lt;/a&gt;
    &lt;/h2&gt;
    &lt;h3&gt;
      The RepoSeance is an AI-powered software archaeology tool designed to help developers understand abandoned, archived, or unfamiliar GitHub repositories.
    &lt;/h3&gt;
  &lt;/div&gt;
  &lt;div class="ltag-github-body"&gt;
    
&lt;div id="readme" class="md"&gt;&lt;p&gt;&lt;a rel="noopener noreferrer" href="https://github.com/Samar-365/RepoSeance/assets/banner.png"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fraw.githubusercontent.com%2FSamar-365%2FRepoSeance%2FHEAD%2Fassets%2Fbanner.png" alt="RepoSeance Banner"&gt;&lt;/a&gt;&lt;/p&gt;
&lt;div class="markdown-heading"&gt;
&lt;h1 class="heading-element"&gt;RepoSeance&lt;/h1&gt;
&lt;/div&gt;
&lt;div class="markdown-heading"&gt;
&lt;h3 class="heading-element"&gt;&lt;em&gt;AI-Powered Software Archaeology &amp;amp; Codebase Resurrection&lt;/em&gt;&lt;/h3&gt;
&lt;/div&gt;
&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Hackathon:&lt;/strong&gt; Hacktoberfest 2026 — &lt;em&gt;Ghost in the Machine&lt;/em&gt;&lt;br&gt;
&lt;strong&gt;Oracle AI:&lt;/strong&gt; Google Gemma 2 (Local Inference via Ollama)&lt;br&gt;
&lt;strong&gt;Tech Stack:&lt;/strong&gt; Python, Flask, React, Vite, SQLite (FTS5), GitPython&lt;/p&gt;
&lt;/blockquote&gt;

&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;Overview&lt;/h2&gt;
&lt;/div&gt;
&lt;p&gt;&lt;strong&gt;RepoSeance&lt;/strong&gt; is an AI-powered software archaeology tool designed to explore, understand, and revive abandoned, archived, or unfamiliar codebases.&lt;/p&gt;
&lt;p&gt;Rather than treating legacy code as disposable, RepoSeance reconstructs a project's development history from its Git commits, source code markers (&lt;code&gt;TODO&lt;/code&gt;, &lt;code&gt;FIXME&lt;/code&gt;, &lt;code&gt;HACK&lt;/code&gt;), unresolved issues, and pull requests. Users interact with a conversational AI assistant that adopts the persona of the &lt;strong&gt;Repository Ghost&lt;/strong&gt; — providing evidence-grounded answers about past architectural decisions, recurring bugs, and unfinished ambitions without hallucinating.&lt;/p&gt;

&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;Key Capabilities&lt;/h2&gt;

&lt;/div&gt;
&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;100% Local Inference &amp;amp; Privacy&lt;/strong&gt;: In local mode, source code and Git metadata never leave your computer. Queries are answered locally using open-weight &lt;strong&gt;Gemma 2&lt;/strong&gt; models.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Strict Evidence Grounding&lt;/strong&gt;: Every claim made…&lt;/li&gt;
&lt;/ul&gt;&lt;/div&gt;
  &lt;/div&gt;
  &lt;div class="gh-btn-container"&gt;&lt;a class="gh-btn" href="https://github.com/Samar-365/RepoSeance" rel="noopener noreferrer"&gt;View on GitHub&lt;/a&gt;&lt;/div&gt;
&lt;/div&gt;





&lt;h2&gt;
  
  
  How I Built It
&lt;/h2&gt;

&lt;p&gt;RepoSeance is built around a &lt;strong&gt;100% open-source, offline AI stack&lt;/strong&gt;:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Frontend: React 19 + Vite (Minimalist Dark Theme &amp;amp; Bespoke SVG Visuals)
        │
        ▼ (REST API)
Backend: Flask Application (Python 3.11)
        │
        ├── Ingestion: GitPython (commit churn, reflog &amp;amp; branch analysis)
        ├── AST Parser: Source scanner for TODO / FIXME / HACK markers
        ├── Storage: SQLite with FTS5 (BM25 full-text search index)
        └── Local AI Engine: Ollama running open-weight Gemma 2:2b
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;

&lt;h3&gt;
  
  
  The Grounding Pipeline
&lt;/h3&gt;

&lt;p&gt;When a user asks: &lt;em&gt;"What was the last major change?"&lt;/em&gt;, RepoSeance does not blindly prompt an LLM:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Local Retrieval&lt;/strong&gt;: It executes a BM25 ranked search across SQLite FTS5 commit tables, commit messages, file diffs, and issue logs.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Context Assembly&lt;/strong&gt;: The top surviving records are injected into a structured grounding prompt.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Local Inference&lt;/strong&gt;: &lt;strong&gt;Gemma 2 (&lt;code&gt;gemma2:2b&lt;/code&gt;)&lt;/strong&gt; processes the grounded context via Ollama with strict system instructions to cite only surviving artifacts.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Citation Extraction&lt;/strong&gt;: The backend parses commit SHAs and file paths from Gemma's output and renders clickable inspector pills in the frontend.&lt;/li&gt;
&lt;/ol&gt;


&lt;h2&gt;
  
  
  Why Does Open Innovation Matter?
&lt;/h2&gt;

&lt;p&gt;Open-source and open-weight AI weren't just nice-to-haves for this project—they were the foundational prerequisite. Here is why open innovation made RepoSeance possible where closed APIs fall short:&lt;/p&gt;
&lt;h3&gt;
  
  
  1. Proprietary Codebases Never Leave Your Machine
&lt;/h3&gt;

&lt;p&gt;Most production repositories contain proprietary business logic, sensitive internal architecture, or compliance-restricted code. Sending your entire git log and diff history to third-party cloud APIs is an immediate security dealbreaker for most engineering teams. Because Gemma 2 runs entirely locally via Ollama, &lt;strong&gt;zero bytes of code or prompt context ever leave your laptop&lt;/strong&gt;.&lt;/p&gt;
&lt;h3&gt;
  
  
  2. Works With the Lights (and the Internet) Off
&lt;/h3&gt;

&lt;p&gt;Software archaeology often happens on airplanes, in server rooms with air-gapped networks, or during outages. RepoSeance has zero external cloud dependencies. You can clone an archived repo into a cabin in the woods with no Wi-Fi, run &lt;code&gt;reposeance&lt;/code&gt;, and interrogate years of engineering history seamlessly.&lt;/p&gt;
&lt;h3&gt;
  
  
  3. Longevity &amp;amp; Zero Marginal Cost
&lt;/h3&gt;

&lt;p&gt;Closed AI services deprecate endpoints, change rate limits, and charge per million tokens. Running a deep audit over a 10,000-commit repository on a metered cloud API could cost tens of dollars per run. With local open-weight Gemma, you can re-index and query codebases indefinitely for &lt;strong&gt;$0.00&lt;/strong&gt;.&lt;/p&gt;


&lt;h2&gt;
  
  
  My Agent Session
&lt;/h2&gt;


&lt;div class="ltag-agent-session"&gt;
  &lt;div class="agent-session-header"&gt;
    
    &lt;span class="agent-session-tool-icon-badge" title="Gemini CLI"&gt;
&lt;/span&gt;
    &lt;span class="agent-session-title"&gt;Building RepoSeance: Local Software Archaeology with Gemma 2&lt;/span&gt;
  &lt;/div&gt;

  &lt;div class="agent-session-scroll"&gt;

      &lt;div class="agent-session-message agent-session-user"&gt;
        &lt;div class="agent-session-role-badge agent-session-role-user"&gt;
          You
        &lt;/div&gt;
        &lt;div class="agent-session-content"&gt;
                &lt;div class="agent-session-text"&gt;
                  &lt;p&gt;I want to build an open-source AI project for the Hacktoberfest 'Ghost in the Machine' challenge called RepoSeance. It should be a software archaeology tool that resuscitates abandoned or legacy codebases by interrogating Git commit history, code markers (TODO/FIXME), and unmerged branches using local open-weight AI (Google Gemma 2) with strict evidence citations.&lt;/p&gt;

                &lt;/div&gt;
        &lt;/div&gt;
      &lt;/div&gt;

      &lt;div class="agent-session-message agent-session-assistant"&gt;
        &lt;div class="agent-session-role-badge agent-session-role-assistant"&gt;
          Agent
        &lt;/div&gt;
        &lt;div class="agent-session-content"&gt;
                &lt;div&gt;
                  &lt;div class="agent-session-text agent-session-text-collapse"&gt;
                    &lt;p&gt;Architecting RepoSeance with local-first, privacy-preserving software archaeology:&lt;br&gt;
1. Backend: Flask with GitPython to traverse commit graphs, detect high-churn inflection points, and unmerged branch divergences.&lt;br&gt;
2. Code Scanner: AST &amp;amp; regex parser for TODO, FIXME, HACK, and DEPRECATED markers categorized by severity.&lt;br&gt;
3. Storage: SQLite with FTS5 BM25 full-text search indexing commit messages and file diffs.&lt;br&gt;
4. AI Inference: Local Ollama runtime with Google's open-weight Gemma 2 (&lt;code&gt;gemma2:2b&lt;/code&gt;) model.&lt;br&gt;
5. Evidence Grounding: Cross-referencing queries with SQLite FTS5 search to cite exact commit SHAs and line numbers without hallucinating.&lt;/p&gt;

                  &lt;/div&gt;
                  Show more
                &lt;/div&gt;
              &lt;div class="agent-session-tool-call"&gt;
                
                  &lt;span class="agent-session-chevron"&gt;▸&lt;/span&gt;
                  &lt;span class="tool-name tool-name-createbackendservice"&gt;create_backend_service&lt;/span&gt;
                    &lt;span class="tool-input-preview"&gt;backend/app.py, backend/indexer.py, backend/ai/ollama_client.py&lt;/span&gt;
                
                &lt;div class="agent-session-tool-detail"&gt;
                    &lt;div class="tool-section"&gt;
                      &lt;div class="tool-section-label"&gt;Input&lt;/div&gt;
                      &lt;pre class="agent-session-pre"&gt;&lt;code&gt;backend/app.py, backend/indexer.py, backend/ai/ollama_client.py&lt;/code&gt;&lt;/pre&gt;
                    &lt;/div&gt;
                    &lt;div class="tool-section"&gt;
                      &lt;div class="tool-section-label"&gt;Output&lt;/div&gt;
                      &lt;pre class="agent-session-pre"&gt;&lt;code&gt;Initialized Flask REST API, SQLite FTS5 search tables, Git collector, and local Gemma 2 Ollama inference client.&lt;/code&gt;&lt;/pre&gt;
                    &lt;/div&gt;
                &lt;/div&gt;
              &lt;/div&gt;
        &lt;/div&gt;
      &lt;/div&gt;

      &lt;div class="agent-session-message agent-session-user"&gt;
        &lt;div class="agent-session-role-badge agent-session-role-user"&gt;
          You
        &lt;/div&gt;
        &lt;div class="agent-session-content"&gt;
                &lt;div class="agent-session-text"&gt;
                  &lt;p&gt;Build the frontend with React 19 and Vite. Make it minimalist, clean, non-gothic with subtle theme line-art visuals, an interactive commit timeline, TODO audit vault, grounded chat with citation drawer, and resurrection report exporter.&lt;/p&gt;

                &lt;/div&gt;
        &lt;/div&gt;
      &lt;/div&gt;

      &lt;div class="agent-session-message agent-session-assistant"&gt;
        &lt;div class="agent-session-role-badge agent-session-role-assistant"&gt;
          Agent
        &lt;/div&gt;
        &lt;div class="agent-session-content"&gt;
                &lt;div class="agent-session-text"&gt;
                  &lt;p&gt;Created modern minimalist interface:&lt;br&gt;
1. Landing View: Hero with minimalist ghost vector and quick repository presets.&lt;br&gt;
2. Grounded Chat: Interactive chat with Gemma 2 citing verified commit SHAs, author timestamps, and diff excerpts.&lt;br&gt;
3. Ghost Timeline: Chronological graph tracking major refactors, milestones, and abandoned branches.&lt;br&gt;
4. Issues &amp;amp; Markers: Excavation audit cataloging TODO/FIXME markers by severity and commit age.&lt;br&gt;
5. Resurrection Report: Automated Markdown report synthesis with feasibility score and revival checklist.&lt;/p&gt;

                &lt;/div&gt;
              &lt;div class="agent-session-tool-call"&gt;
                
                  &lt;span class="agent-session-chevron"&gt;▸&lt;/span&gt;
                  &lt;span class="tool-name tool-name-buildfrontendviews"&gt;build_frontend_views&lt;/span&gt;
                    &lt;span class="tool-input-preview"&gt;frontend/src/views/LandingView.jsx, frontend/src/views/SeanceChatView.jsx, fr...&lt;/span&gt;
                
                &lt;div class="agent-session-tool-detail"&gt;
                    &lt;div class="tool-section"&gt;
                      &lt;div class="tool-section-label"&gt;Input&lt;/div&gt;
                      &lt;pre class="agent-session-pre"&gt;&lt;code&gt;frontend/src/views/LandingView.jsx, frontend/src/views/SeanceChatView.jsx, frontend/src/views/GhostTimelineView.jsx&lt;/code&gt;&lt;/pre&gt;
                    &lt;/div&gt;
                    &lt;div class="tool-section"&gt;
                      &lt;div class="tool-section-label"&gt;Output&lt;/div&gt;
                      &lt;pre class="agent-session-pre"&gt;&lt;code&gt;React 19 views built and verified with Vite HMR at http://localhost:5173.&lt;/code&gt;&lt;/pre&gt;
                    &lt;/div&gt;
                &lt;/div&gt;
              &lt;/div&gt;
        &lt;/div&gt;
      &lt;/div&gt;

      &lt;div class="agent-session-message agent-session-user"&gt;
        &lt;div class="agent-session-role-badge agent-session-role-user"&gt;
          You
        &lt;/div&gt;
        &lt;div class="agent-session-content"&gt;
                &lt;div class="agent-session-text"&gt;
                  &lt;p&gt;Run end-to-end integration tests to verify the SQLite FTS5 search, Ollama Gemma 2 citation grounding, and repository analysis pipeline.&lt;/p&gt;

                &lt;/div&gt;
        &lt;/div&gt;
      &lt;/div&gt;

      &lt;div class="agent-session-message agent-session-assistant"&gt;
        &lt;div class="agent-session-role-badge agent-session-role-assistant"&gt;
          Agent
        &lt;/div&gt;
        &lt;div class="agent-session-content"&gt;
                &lt;div class="agent-session-text"&gt;
                  &lt;p&gt;Executed 10-step end-to-end integration test suite:&lt;br&gt;
- Health check: OK (Ollama connected with gemma2:2b)&lt;br&gt;
- Repository ingestion: OK (Analyzed commits, authors, and file markers)&lt;br&gt;
- FTS5 BM25 search: OK (Matched historical terms and commit SHAs)&lt;br&gt;
- Chat grounding: OK (Responses cite surviving records without hallucinating)&lt;br&gt;
- Resurrection report: OK (Markdown audit and feasibility metrics generated)&lt;/p&gt;

                &lt;/div&gt;
              &lt;div class="agent-session-tool-call"&gt;
                
                  &lt;span class="agent-session-chevron"&gt;▸&lt;/span&gt;
                  &lt;span class="tool-name tool-name-runintegrationtests"&gt;run_integration_tests&lt;/span&gt;
                    &lt;span class="tool-input-preview"&gt;python backend/test_e2e.py&lt;/span&gt;
                
                &lt;div class="agent-session-tool-detail"&gt;
                    &lt;div class="tool-section"&gt;
                      &lt;div class="tool-section-label"&gt;Input&lt;/div&gt;
                      &lt;pre class="agent-session-pre"&gt;&lt;code&gt;python backend/test_e2e.py&lt;/code&gt;&lt;/pre&gt;
                    &lt;/div&gt;
                    &lt;div class="tool-section"&gt;
                      &lt;div class="tool-section-label"&gt;Output&lt;/div&gt;
                      &lt;pre class="agent-session-pre"&gt;&lt;code&gt;10/10 tests passed successfully.&lt;/code&gt;&lt;/pre&gt;
                    &lt;/div&gt;
                &lt;/div&gt;
              &lt;/div&gt;
        &lt;/div&gt;
      &lt;/div&gt;
  &lt;/div&gt;

  &lt;div class="agent-session-footer"&gt;
    &lt;span class="agent-session-meta"&gt;
        6 of 6 messages
    &lt;/span&gt;
  &lt;/div&gt;
&lt;/div&gt;



&lt;p&gt;&lt;em&gt;Full session transcript published with DevRelay: &lt;a href="https://dev.to/agent_sessions/building-reposeance-local-software-archaeology-with-gemma-2-kur6xe"&gt;View Agent Session&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;




&lt;h2&gt;
  
  
  The Spirit of It
&lt;/h2&gt;

&lt;p&gt;In Mexican tradition, &lt;em&gt;Día de los Muertos&lt;/em&gt; is not a day of mourning, but an act of remembrance—a belief that someone only truly disappears when their stories are no longer told. &lt;/p&gt;

&lt;p&gt;Software development has an unfortunate habit of treating past code with cynicism. We scoff at legacy systems, call older code "spaghetti", and rewrite systems from scratch without understanding the hard-learned lessons that shaped them.&lt;/p&gt;

&lt;p&gt;RepoSeance was built to honor the past. It gives voice to the developers who wrote code at 2 AM five years ago to keep a service alive, preserves their context, and treats their unfinished ambitions with respect. It turns a graveyard of dead commits into an ofrenda of engineering wisdom.&lt;/p&gt;




&lt;h2&gt;
  
  
  Prize Categories
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Featured Category: Best Use of Gemma ($200)&lt;/strong&gt;: Powered locally by Google's open-weight &lt;code&gt;gemma2:2b&lt;/code&gt; model for grounding, architectural Q&amp;amp;A, and resurrection report synthesis.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Partner Category: Best Use of Entire ($100)&lt;/strong&gt;: Full agent session transcript published and embedded via DevRelay (&lt;code&gt;{% agent_session 1320 %}&lt;/code&gt;).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Top Prize&lt;/strong&gt;: Complete submission for Hacktoberfest Week 4 &lt;em&gt;Ghost in the Machine&lt;/em&gt;.&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
    </item>
    <item>
      <title>ReMind: The Sloth-Mode Developer Resumption Assistant Powered by Local Gemma 2</title>
      <dc:creator>Samar Shetye</dc:creator>
      <pubDate>Sun, 25 Oct 2026 14:31:14 +0000</pubDate>
      <link>https://dev.to/samar_shetye/remind-the-sloth-mode-developer-resumption-assistant-powered-by-local-gemma-2-33ed</link>
      <guid>https://dev.to/samar_shetye/remind-the-sloth-mode-developer-resumption-assistant-powered-by-local-gemma-2-33ed</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-week3-2026-10-19"&gt;Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What I Built
&lt;/h2&gt;

&lt;p&gt;Every developer knows the dreaded &lt;strong&gt;"Morning Friction"&lt;/strong&gt; penalty. &lt;br&gt;
You finish a late-night hacking session, close your laptop without writing documentation, and go to sleep. The next morning, you spend 25 minutes staring at your terminal trying to reconstruct your train of thought:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;em&gt;What branch was I on?&lt;/em&gt;&lt;/li&gt;
&lt;li&gt;&lt;em&gt;Why did I leave this git diff uncommitted?&lt;/em&gt;&lt;/li&gt;
&lt;li&gt;&lt;em&gt;What was that StackOverflow tab I had open at 1:00 AM?&lt;/em&gt;&lt;/li&gt;
&lt;li&gt;
&lt;em&gt;What broken test was I about to fix?&lt;/em&gt;
Writing daily handoff notes is the responsible thing to do, but at the end of the day, you're exhausted and want to be a sloth.
&lt;strong&gt;ReMind&lt;/strong&gt; is a local-by-default, privacy-preserving developer resumption daemon and dashboard. It allows you to be totally lazy: close your laptop without writing a single line of notes. 
In the background, ReMind automatically aggregates your day's work across:&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Multi-Project Git Activity&lt;/strong&gt;: Commit logs, branch states, and unstaged diffs across all repositories you touched.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Terminal History&lt;/strong&gt;: Recent PowerShell and Bash/Zsh command history.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Active Desktop Windows&lt;/strong&gt;: Open IDE files, devtools, and editor windows you were looking at right before signing off.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Browser Research Tabs&lt;/strong&gt;: Local Chrome/Edge/Brave history to identify technical docs and discussions you referenced.
It scrubs all secrets and passes the sanitized context to &lt;strong&gt;Google's Gemma 2&lt;/strong&gt; running locally on your machine via &lt;strong&gt;Ollama&lt;/strong&gt;. In seconds, it produces a structured, actionable &lt;strong&gt;Tomorrow-Me Resumption Brief&lt;/strong&gt; complete with:&lt;/li&gt;
&lt;li&gt;A plain-English task summary&lt;/li&gt;
&lt;li&gt;Evidence-based roadblock hypotheses&lt;/li&gt;
&lt;li&gt;Interactive, clickable next-step checkboxes&lt;/li&gt;
&lt;li&gt;Click-to-copy terminal commands to immediately resume your dev servers&lt;/li&gt;
&lt;/ul&gt;


&lt;h2&gt;
  
  
  Demo
&lt;/h2&gt;

&lt;p&gt;Live App:&lt;/p&gt;


&lt;div class="crayons-card c-embed text-styles text-styles--secondary"&gt;
    &lt;div class="c-embed__content"&gt;
      &lt;div class="c-embed__body flex items-center justify-between"&gt;
        &lt;a href="https://remind-dashboard-e80s.onrender.com" rel="noopener noreferrer" class="c-link fw-bold flex items-center"&gt;
          &lt;span class="mr-2"&gt;remind-dashboard-e80s.onrender.com&lt;/span&gt;
          

        &lt;/a&gt;
      &lt;/div&gt;
    &lt;/div&gt;
&lt;/div&gt;


&lt;p&gt;Youtube Demo Video:&lt;/p&gt;

&lt;p&gt;  &lt;iframe src="https://www.youtube.com/embed/er1CiX3Mv9Y" width="710" height="399"&gt;
  &lt;/iframe&gt;
&lt;/p&gt;

&lt;p&gt;Here is the ReMind workflow in action:&lt;/p&gt;

&lt;h3&gt;
  
  
  1. Multi-Project Daily Activity Scanner
&lt;/h3&gt;

&lt;p&gt;ReMind detects activity across all active repositories, uncommitted file changes, 25 terminal commands, and 5 active windows:&lt;br&gt;
&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fs13pe83j7mzeg3ll9vvr.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fs13pe83j7mzeg3ll9vvr.png" alt="Today's Development Activity" width="800" height="379"&gt;&lt;/a&gt;&lt;/p&gt;
&lt;h3&gt;
  
  
  2. The Gemma 2 Tomorrow-Me Resumption Brief
&lt;/h3&gt;

&lt;p&gt;In one click, Gemma 2 synthesizes the chaotic context into a calm, structured game plan with interactive next steps and modified file tags:&lt;br&gt;
&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F0jie0j0bo2wrwoj0vr3b.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F0jie0j0bo2wrwoj0vr3b.png" alt="Tomorrow's Resumption Brief" width="800" height="380"&gt;&lt;/a&gt;&lt;/p&gt;
&lt;h3&gt;
  
  
  3. Context &amp;amp; Sanitization Inspector
&lt;/h3&gt;

&lt;p&gt;Transparency is key. Before inference, you can inspect the exact terminal commands, git diffs, and window titles captured:&lt;br&gt;
&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fue5jbw57689kkjig9jku.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fue5jbw57689kkjig9jku.png" alt="Context Inspector" width="800" height="378"&gt;&lt;/a&gt;&lt;/p&gt;
&lt;h3&gt;
  
  
  4. Past Handoff History
&lt;/h3&gt;

&lt;p&gt;Searchable chronological archive of previous days' sessions to track project velocity over time:&lt;br&gt;
&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F9l8n4nr10docgjs47o18.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F9l8n4nr10docgjs47o18.png" alt="Past Handoff Notes" width="800" height="377"&gt;&lt;/a&gt;&lt;/p&gt;


&lt;h2&gt;
  
  
  Code
&lt;/h2&gt;

&lt;p&gt;The complete source code is open source under the MIT License:&lt;/p&gt;


&lt;div class="ltag-github-readme-tag"&gt;
  &lt;div class="readme-overview"&gt;
    &lt;h2&gt;
      &lt;img src="https://assets.dev.to/assets/github-logo-5a155e1f9a670af7944dd5e12375bc76ed542ea80224905ecaf878b9157cdefc.svg" alt="GitHub logo"&gt;
      &lt;a href="https://github.com/Samar-365" rel="noopener noreferrer"&gt;
        Samar-365
      &lt;/a&gt; / &lt;a href="https://github.com/Samar-365/ReMind" rel="noopener noreferrer"&gt;
        ReMind
      &lt;/a&gt;
    &lt;/h2&gt;
    &lt;h3&gt;
      Tomorrow-Me Handoff (ReMind)_ A privacy-first, local-by-default AI assistant that captures a developer's end-of-day workflow context (Git diffs, shell commands, browser tabs) and synthesizes a structured resumption briefing using open-weight models (Gemma via Ollama).
    &lt;/h3&gt;
  &lt;/div&gt;
  &lt;div class="ltag-github-body"&gt;
    
&lt;div id="readme" class="md"&gt;&lt;div class="markdown-heading"&gt;
&lt;h1 class="heading-element"&gt;Tomorrow-Me Handoff (ReMind)&lt;/h1&gt;
&lt;/div&gt;
&lt;p&gt;
  &lt;a rel="noopener noreferrer" href="https://github.com/Samar-365/ReMind/./docs/cover.jpg"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fraw.githubusercontent.com%2FSamar-365%2FReMind%2FHEAD%2F.%2Fdocs%2Fcover.jpg" alt="ReMind - Tomorrow-Me Developer Resumption Assistant" width="100%"&gt;&lt;/a&gt;
&lt;/p&gt;

&lt;p&gt;&lt;a href="https://fastapi.tiangolo.com" rel="nofollow noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/5dd6f11e2debf67f83efbdbd92e1386670981d37ae7d6a82c36f78b274bd7385/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f466173744150492d302e3131302b2d3030393638382e7376673f6c6f676f3d66617374617069266c6f676f436f6c6f723d7768697465" alt="FastAPI"&gt;&lt;/a&gt;
&lt;a href="https://react.dev" rel="nofollow noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/22e8e2c0cac5ae42371ca1ffecd0f5246ebf4b26a8a4375b90ae75a5accd517f/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f52656163742d31382b2d3631444146422e7376673f6c6f676f3d7265616374266c6f676f436f6c6f723d626c61636b" alt="React"&gt;&lt;/a&gt;
&lt;a href="https://vitejs.dev" rel="nofollow noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/568e126748f329757e9685c6537db526f65fa0b8473f7319ad8afcca3857bdfa/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f566974652d6c61746573742d3634364346462e7376673f6c6f676f3d76697465266c6f676f436f6c6f723d7768697465" alt="Vite"&gt;&lt;/a&gt;
&lt;a href="https://ollama.ai" rel="nofollow noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/5b6ce3c32744637f3a72604480837baa6afca853fa20dbba4f20c4232aa2a1b3/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f4f6c6c616d612d4c6f63616c5f496e666572656e63652d3030303030302e7376673f6c6f676f3d6f6c6c616d61266c6f676f436f6c6f723d7768697465" alt="Ollama"&gt;&lt;/a&gt;
&lt;a href="https://python.org" rel="nofollow noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/a32434c66a6218d5c449494027a09b4172006dfef226e80e0bc7735ec9f29790/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f507974686f6e2d332e31312b2d3337373641422e7376673f6c6f676f3d707974686f6e266c6f676f436f6c6f723d7768697465" alt="Python"&gt;&lt;/a&gt;
&lt;a href="https://opensource.org/licenses/MIT" rel="nofollow noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/fdf2982b9f5d7489dcf44570e714e3a15fce6253e0cc6b5aa61a075aac2ff71b/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f4c6963656e73652d4d49542d79656c6c6f772e737667" alt="License: MIT"&gt;&lt;/a&gt;&lt;/p&gt;
&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Stop losing morning momentum.&lt;/strong&gt;&lt;br&gt;
An intelligent, privacy-first developer daemon and dashboard that inspects your end-of-day Git diffs, terminal history, and research tabs to synthesize a grounded 7-section resumption brief with local Gemma 2 AI. Zero cloud leakage.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;Key Features&lt;/h2&gt;
&lt;/div&gt;
&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Local-First Privacy &amp;amp; Zero Cloud Leak&lt;/strong&gt;
&lt;ul&gt;
&lt;li&gt;All inference runs directly on your hardware via local &lt;a href="https://ollama.ai/" rel="nofollow noopener noreferrer"&gt;Ollama&lt;/a&gt; (&lt;code&gt;gemma2:2b&lt;/code&gt; or &lt;code&gt;gemma2:9b&lt;/code&gt;).&lt;/li&gt;
&lt;li&gt;No source code, diffs, terminal history, or window titles leave your machine.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Built-in Secret Scrubber &amp;amp; Token Budgeter&lt;/strong&gt;
&lt;ul&gt;
&lt;li&gt;Automatically identifies and masks API keys (OpenAI, AWS, GitHub PATs, private keys, database connection strings, and OAuth tokens) across diffs, terminal commands, browser tabs, and desktop window titles.&lt;/li&gt;
&lt;li&gt;Intelligently truncates large diffs to avoid context window blowup while prioritizing high-signal application files over lockfiles.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Multi-Source Context Aggregation&lt;/strong&gt;
&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Multi-Project Git Discovery&lt;/strong&gt;: Automatically discovers &lt;strong&gt;all git repositories&lt;/strong&gt; you worked on throughout the day, compiling today's commit logs…&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;/ul&gt;&lt;/div&gt;
  &lt;/div&gt;
  &lt;div class="gh-btn-container"&gt;&lt;a class="gh-btn" href="https://github.com/Samar-365/ReMind" rel="noopener noreferrer"&gt;View on GitHub&lt;/a&gt;&lt;/div&gt;
&lt;/div&gt;





&lt;h2&gt;
  
  
  How I Built It
&lt;/h2&gt;

&lt;p&gt;ReMind is built from the ground up around &lt;strong&gt;Google's Gemma 2 (&lt;code&gt;gemma2:2b&lt;/code&gt;)&lt;/strong&gt; running locally through &lt;strong&gt;Ollama&lt;/strong&gt;.&lt;/p&gt;

&lt;h3&gt;
  
  
  The Pipeline Architecture:
&lt;/h3&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;[ Git Diffs &amp;amp; Commits ] [ Terminal History ] ──► [ Secret Scrubber ] ──► [ Gemma 2 (Local Ollama) ] ──► [ Structured Brief ] [ Browser History ] (Masks Keys/Tokens) (Typed Pydantic Schema) (Checklist + Commands) [ Open Desktop Windows]
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Context Harvesting Daemon (&lt;code&gt;context_service.py&lt;/code&gt;)&lt;/strong&gt;:

&lt;ul&gt;
&lt;li&gt;Discovers all active git repositories on your drive by scanning commit timestamps.&lt;/li&gt;
&lt;li&gt;Extracts shell commands from PowerShell (&lt;code&gt;ConsoleHost_history.txt&lt;/code&gt;) and Bash.&lt;/li&gt;
&lt;li&gt;Inspects desktop window titles via Windows API to record open files and tools without recording screen pixels.&lt;/li&gt;
&lt;li&gt;Reads local Chromium SQLite history for recent technical domains (StackOverflow, GitHub, official docs).&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Regex Secret Scrubber &amp;amp; Token Budgeter (&lt;code&gt;sanitizer.py&lt;/code&gt;)&lt;/strong&gt;:

&lt;ul&gt;
&lt;li&gt;Protects sensitive credentials by auto-redacting OpenAI API keys, GitHub PATs, AWS secrets, bearer tokens, and private SSH keys across all inputs.&lt;/li&gt;
&lt;li&gt;Prioritizes high-signal application code over lockfiles (&lt;code&gt;package-lock.json&lt;/code&gt;, &lt;code&gt;poetry.lock&lt;/code&gt;) to stay within Gemma's context window.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Gemma 2 Prompt Engineering &amp;amp; Structured Inference (&lt;code&gt;ai.py&lt;/code&gt;)&lt;/strong&gt;:

&lt;ul&gt;
&lt;li&gt;Uses strict system prompts and Pydantic JSON schemas to constrain Gemma 2 to produce clean, anti-hallucinatory resumption structures: &lt;code&gt;current_task&lt;/code&gt;, &lt;code&gt;progress_summary&lt;/code&gt;, &lt;code&gt;roadblocks_and_hypotheses&lt;/code&gt;, and &lt;code&gt;recommended_next_steps&lt;/code&gt;.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Web Dashboard &amp;amp; Background Scheduler (&lt;code&gt;frontend/&lt;/code&gt; &amp;amp; &lt;code&gt;scheduler.py&lt;/code&gt;)&lt;/strong&gt;:

&lt;ul&gt;
&lt;li&gt;An end-of-day timer triggers automatically when you log off, sending a native desktop toast notification when tomorrow's briefing is compiled.&lt;/li&gt;
&lt;li&gt;The React frontend provides an interactive checklist so you can cross off morning tasks as you complete them.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;/ol&gt;




&lt;h2&gt;
  
  
  Why Does Open Innovation Matter?
&lt;/h2&gt;

&lt;p&gt;This project simply &lt;strong&gt;could not exist with closed, cloud-based LLM APIs&lt;/strong&gt;. Here is why:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Proprietary Source Code Never Leaves Your Device&lt;/strong&gt;:
Developers and enterprise organizations cannot pipe uncommitted git diffs, internal repository names, and private terminal commands to closed cloud APIs like OpenAI or Anthropic without violating privacy and IP policies. Because &lt;strong&gt;Gemma 2 is an open-weight model&lt;/strong&gt;, all inference runs entirely in-memory on your local CPU/GPU.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Zero Operating Cost&lt;/strong&gt;:
Running a daily automated daemon using commercial API tokens would incur monthly recurring bills. Open-weight models like Gemma cost $0 to run indefinitely.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;100% Offline Capability&lt;/strong&gt;:
Whether you are working on a flight, in a café with spotty Wi-Fi, or in an air-gapped environment, ReMind works seamlessly without internet access.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Model Swapping &amp;amp; Freedom&lt;/strong&gt;:
Because ReMind integrates with standard open-inference protocols, users can switch between &lt;code&gt;gemma2:2b&lt;/code&gt; for lightweight ultra-fast laptop inference, or scale up to &lt;code&gt;gemma2:9b&lt;/code&gt; or fine-tuned coding weights with zero code changes.&lt;/li&gt;
&lt;/ol&gt;




&lt;h2&gt;
  
  
  The Lazy Payoff
&lt;/h2&gt;

&lt;p&gt;International Sloth Day reminds us that efficiency isn't about working harder—it's about eliminating unnecessary friction. &lt;br&gt;
Before ReMind, resuming a multi-repo project took an average of &lt;strong&gt;15 to 25 minutes of cognitive overhead&lt;/strong&gt; every morning:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Retracing terminal history&lt;/li&gt;
&lt;li&gt;Reading &lt;code&gt;git status&lt;/code&gt; across multiple folders&lt;/li&gt;
&lt;li&gt;Re-orienting mental models
With ReMind:&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Time saved&lt;/strong&gt;: ~20 minutes every single morning (~1.5 hours per week).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Effort saved&lt;/strong&gt;: You never have to write end-of-day status summaries, slack updates, or standup notes manually again. Open your dashboard, review Gemma's checklist, copy your standup notes in 1 click, and start coding.&lt;/li&gt;
&lt;/ul&gt;




&lt;h2&gt;
  
  
  Prize Categories
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Featured Categories:
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Best Use of Gemma ($200)&lt;/strong&gt;: ReMind is built entirely around Google's open-weight &lt;code&gt;gemma2:2b&lt;/code&gt; model, utilizing local Ollama inference, structured prompt schemas, and on-device privacy guarantees.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Best Use of Render ($200)&lt;/strong&gt;: 
The ReMind web dashboard and landing page are hosted directly on &lt;strong&gt;Render&lt;/strong&gt; using an Infrastructure-as-Code Blueprint (&lt;a href="https://github.com/Samar-365/ReMind/blob/main/render.yaml" rel="noopener noreferrer"&gt;&lt;code&gt;render.yaml&lt;/code&gt;&lt;/a&gt;). Render automates the continuous deployment of our React/Vite frontend as a global CDN static site, providing an interactive online showcase for developers to explore the resumption interface. 
&lt;em&gt;(Live Deployment: &lt;a href="https://remind-dashboard-e80s.onrender.com" rel="noopener noreferrer"&gt;remind-dashboard on Render&lt;/a&gt;)&lt;/em&gt;
&lt;/li&gt;
&lt;/ul&gt;




&lt;p&gt;&lt;em&gt;Thank you to DEV, GitHub, and Google for organizing Hacktoberfest 2026 Week 3!&lt;/em&gt;&lt;/p&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
    </item>
    <item>
      <title>JobFit: A 100% Local-First AI Career Copilot That Kills Resume Hallucinations</title>
      <dc:creator>Samar Shetye</dc:creator>
      <pubDate>Sun, 18 Oct 2026 13:49:01 +0000</pubDate>
      <link>https://dev.to/samar_shetye/jobfit-a-100-local-first-ai-career-copilot-that-kills-resume-hallucinations-37in</link>
      <guid>https://dev.to/samar_shetye/jobfit-a-100-local-first-ai-career-copilot-that-kills-resume-hallucinations-37in</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-week2-2026-10-12"&gt;Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What I Built
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;JobFit&lt;/strong&gt; is an ethical, local-first AI career application copilot designed to empower job seekers with &lt;strong&gt;verifiable, evidence-backed job matching&lt;/strong&gt;, &lt;strong&gt;grounded resume bullet tailoring&lt;/strong&gt;, and &lt;strong&gt;truthful cover letter drafting&lt;/strong&gt; — operating with &lt;strong&gt;zero cloud dependencies&lt;/strong&gt; and &lt;strong&gt;zero external data transmission&lt;/strong&gt;.&lt;/p&gt;

&lt;h3&gt;
  
  
  The Problem It Solves
&lt;/h3&gt;

&lt;p&gt;Job seekers and career consultants today face two major hazards with commercial cloud AI tools:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Severe Privacy Breaches&lt;/strong&gt;: Candidates upload lifetime employment histories, contact info, and addresses to cloud wrappers that log and train on sensitive personal records.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Hallucinated "ATS Scores" &amp;amp; Fabricated Achievements&lt;/strong&gt;: Commercial matchers invent opaque percentage scores without citing proof. Worse, AI rephrasers frequently fabricate metrics (e.g. turning &lt;em&gt;"built internal portal"&lt;/em&gt; into &lt;em&gt;"scaled distributed architecture by 450% generating $12M"&lt;/em&gt;), putting candidates at legal risk during background checks.&lt;/li&gt;
&lt;/ol&gt;

&lt;h3&gt;
  
  
  Who Would Pay For It? (The Side Hustle Angle)
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Privacy-Conscious Job Seekers &amp;amp; Tech Professionals&lt;/strong&gt;: Individuals applying for high-trust roles (finance, defense, tech) who cannot leak company IP or personal PII to cloud LLMs.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Independent Resume Coaches &amp;amp; Career Consultants&lt;/strong&gt;: A professional career consultancy running JobFit can offer verified, hallucination-free resume optimization services to clients with &lt;strong&gt;zero per-token OpenAI API costs&lt;/strong&gt;, keeping 100% of the profit margins.&lt;/li&gt;
&lt;/ul&gt;




&lt;h2&gt;
  
  
  Demo
&lt;/h2&gt;

&lt;p&gt;Youtube Video:&lt;/p&gt;

&lt;p&gt;  &lt;iframe src="https://www.youtube.com/embed/WTwe2BYErfA" width="710" height="399"&gt;
  &lt;/iframe&gt;
&lt;/p&gt;

&lt;h3&gt;
  
  
  Highlights in the Demo:
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Multi-Format Ingestion&lt;/strong&gt;: Ingesting PDF, DOCX, and TXT resumes directly into structured candidate profiles.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Evidence-Backed Matching&lt;/strong&gt;: Transparent mathematical breakdown citing verbatim resume lines for every match.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Anti-Fabrication Guardrail&lt;/strong&gt;: Real-time regex inspection preventing invented numbers or credentials.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Application Tracker&lt;/strong&gt;: Pipeline management with 1-click CSV and Markdown evidence report exports.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;1-Click Local Data Purge&lt;/strong&gt;: Full data sovereignty with instant SQLite and file cleanup.&lt;/li&gt;
&lt;/ul&gt;




&lt;h2&gt;
  
  
  Code
&lt;/h2&gt;

&lt;p&gt;Github Repository:&lt;/p&gt;


&lt;div class="ltag-github-readme-tag"&gt;
  &lt;div class="readme-overview"&gt;
    &lt;h2&gt;
      &lt;img src="https://assets.dev.to/assets/github-logo-5a155e1f9a670af7944dd5e12375bc76ed542ea80224905ecaf878b9157cdefc.svg" alt="GitHub logo"&gt;
      &lt;a href="https://github.com/Samar-365" rel="noopener noreferrer"&gt;
        Samar-365
      &lt;/a&gt; / &lt;a href="https://github.com/Samar-365/jobfit" rel="noopener noreferrer"&gt;
        jobfit
      &lt;/a&gt;
    &lt;/h2&gt;
    &lt;h3&gt;
      
    &lt;/h3&gt;
  &lt;/div&gt;
  &lt;div class="ltag-github-body"&gt;
    
&lt;div id="readme" class="md"&gt;&lt;div class="markdown-heading"&gt;
&lt;h1 class="heading-element"&gt;JobFit — Local-First AI Job Application Assistant&lt;/h1&gt;
&lt;/div&gt;
&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;JobFit&lt;/strong&gt; is an ethical, local-first AI career application copilot designed to empower candidates with &lt;strong&gt;verifiable, evidence-backed job matching&lt;/strong&gt;, &lt;strong&gt;grounded resume bullet tailoring&lt;/strong&gt;, and &lt;strong&gt;truthful cover letter drafting&lt;/strong&gt; — operating with &lt;strong&gt;zero cloud dependencies&lt;/strong&gt; and &lt;strong&gt;zero external data transmission&lt;/strong&gt;.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;The Problem with Cloud Career Tools&lt;/h2&gt;
&lt;/div&gt;
&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Severe Privacy Hazards&lt;/strong&gt;: Candidates upload their lifetime work histories, contact numbers, addresses, and compensation notes to third-party cloud LLM wrappers that log and train on sensitive personal records.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Hallucinated "ATS Scores"&lt;/strong&gt;: Commercial resume matchers use black-box scoring algorithms that invent scores (e.g., "78% ATS Match") without explaining what is missing or citing proof.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Fabricated Credentials&lt;/strong&gt;: Cloud AI rephrasers frequently invent numbers (e.g., changing &lt;em&gt;"built internal portal"&lt;/em&gt; into &lt;em&gt;"scaled distributed architecture by 450% generating $12M"&lt;/em&gt;), putting candidates at legal and ethical risk during background checks.&lt;/li&gt;
&lt;/ol&gt;

&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;The JobFit Solution&lt;/h2&gt;
&lt;/div&gt;
&lt;p&gt;JobFit solves these problems by…&lt;/p&gt;&lt;/div&gt;
  &lt;/div&gt;
  &lt;div class="gh-btn-container"&gt;&lt;a class="gh-btn" href="https://github.com/Samar-365/jobfit" rel="noopener noreferrer"&gt;View on GitHub&lt;/a&gt;&lt;/div&gt;
&lt;/div&gt;





&lt;h2&gt;
  
  
  How I Built It
&lt;/h2&gt;

&lt;p&gt;JobFit is engineered as an entirely self-contained workstation application: &lt;/p&gt;

&lt;h3&gt;
  
  
  The Open-Source AI Engine
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Google Gemma 2 (2B)&lt;/strong&gt;: Quantized (&lt;code&gt;Q4_K_M&lt;/code&gt;, ~1.6 GB) running locally via &lt;strong&gt;Ollama&lt;/strong&gt;. It delivers sub-second to 15s inference directly on consumer GPUs (NVIDIA RTX 4050 Laptop) or modern CPUs with zero cloud API keys.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Fenced XML Prompt Boundaries&lt;/strong&gt;: User-supplied resumes and job postings are strictly isolated within &lt;code&gt;&amp;lt;resume_document&amp;gt;&lt;/code&gt; and &lt;code&gt;&amp;lt;job_posting&amp;gt;&lt;/code&gt; XML delimiters to prevent prompt injection and adversarial system overrides.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Verbatim Citation Extraction&lt;/strong&gt;: Gemma is prompted to act as an objective auditor, required to extract exact quotes for matches or classify criteria as &lt;code&gt;NOT_MENTIONED&lt;/code&gt; rather than hallucinating shortcomings.&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  The Application Stack
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Backend&lt;/strong&gt;: Python 3.11, &lt;strong&gt;FastAPI&lt;/strong&gt;, SQLite, SQLAlchemy 2.0, &lt;code&gt;pypdf&lt;/code&gt;, &lt;code&gt;python-docx&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Frontend&lt;/strong&gt;: &lt;strong&gt;React 18&lt;/strong&gt;, TypeScript, Vite 6, styled with a custom &lt;strong&gt;Pastel 3D Studio Design System&lt;/strong&gt; (tactile 3D cards, porcelain off-white canvas &lt;code&gt;#F8F7F4&lt;/code&gt;, soft pastel accents).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Deterministic Scoring Engine&lt;/strong&gt;: Replaces black-box heuristics with an open formula:
$$\text{Score} = \left(\frac{\text{Weighted Verified Points}}{\text{Total Criteria}}\right) \times 100$$&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Programmatic Guardrail Validator&lt;/strong&gt;: Inspects LLM bullet suggestions against the candidate's original text to mathematically reject any hallucinated metrics or percentages.&lt;/li&gt;
&lt;/ul&gt;




&lt;h2&gt;
  
  
  Why Does Open Innovation Matter?
&lt;/h2&gt;

&lt;p&gt;Open-weight models like &lt;strong&gt;Google Gemma 2&lt;/strong&gt; make applications like JobFit possible in ways closed commercial APIs (OpenAI, Anthropic) cannot:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Zero Data Transmission&lt;/strong&gt;: Career data contains the most intimate summary of a person's life. Open-source local inference is the only guarantee that a candidate's personal data never leaves their physical machine.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Predictable Unit Economics&lt;/strong&gt;: A freelance career consultant can run millions of evaluation tokens locally through Ollama with &lt;strong&gt;$0 in API bills&lt;/strong&gt;, turning an AI side hustle into a sustainable, high-margin business.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;No Vendor Lock-In or Sudden Deprecations&lt;/strong&gt;: The entire stack is self-hosted, reproducible, and verifiable.&lt;/li&gt;
&lt;/ol&gt;




&lt;h2&gt;
  
  
  My First User
&lt;/h2&gt;

&lt;p&gt;To test JobFit in the real world, I shared the build with a fellow software engineering friend who has been actively applying for roles in the current tough tech market.&lt;/p&gt;

&lt;h3&gt;
  
  
  Their Experience &amp;amp; Feedback:
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;What they tested&lt;/strong&gt;: They uploaded their real resume (Alex Rivera archetype) and ran it against a Senior Fullstack React/Python job posting.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;What impressed them most&lt;/strong&gt;: 

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Honest Citations&lt;/strong&gt;: They were amazed that every single requirement match cited the exact verbatim sentence from their resume instead of giving an arbitrary black-box score like commercial tools.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;No Hallucinated Lies&lt;/strong&gt;: In their words: &lt;em&gt;"ChatGPT always turns my simple API projects into 'scaled enterprise infrastructure generating $5M'. JobFit's tailoring kept my numbers accurate so I won't get embarrassed in a technical interview."&lt;/em&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Privacy Peace of Mind&lt;/strong&gt;: They loved knowing their phone number, email, and personal background never touched a cloud server.&lt;/li&gt;
&lt;/ol&gt;
&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  Would they pay?
&lt;/h3&gt;

&lt;p&gt;&lt;strong&gt;Yes.&lt;/strong&gt; They mentioned they currently pay $20/month for a popular cloud resume scanner that hallucinates constantly. They said they would readily pay a &lt;strong&gt;$15–$25 one-time lifetime license&lt;/strong&gt; (or $10/month) for JobFit as an offline desktop application because of the zero-cloud privacy guarantee and honest evidence matching.&lt;/p&gt;

&lt;h3&gt;
  
  
  What they asked for next:
&lt;/h3&gt;

&lt;ol&gt;
&lt;li&gt;One-click PDF export for the tailored resume.&lt;/li&gt;
&lt;li&gt;A direct URL scraper to paste a LinkedIn or Indeed job link instead of copying and pasting the job text.&lt;/li&gt;
&lt;/ol&gt;




&lt;h2&gt;
  
  
  Prize Categories
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Best Use of Gemma ($200)
&lt;/h3&gt;

&lt;p&gt;JobFit is built natively around &lt;strong&gt;Google Gemma 2 (2B)&lt;/strong&gt; (&lt;code&gt;Q4_K_M&lt;/code&gt; quantization) running 100% locally on the candidate's workstation via Ollama:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Local Sovereignty&lt;/strong&gt;: Runs on &lt;code&gt;localhost:11434&lt;/code&gt; with zero external cloud dependencies or telemetry, ensuring sensitive personal career history never leaves the user's laptop.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Evidence-Backed Fact Extraction&lt;/strong&gt;: Gemma acts as an objective evaluator, extracting exact verbatim quotes from submitted resumes to verify qualifications rather than inventing arbitrary "ATS match scores".&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Prompt Injection Defense&lt;/strong&gt;: Gemma's inference is encapsulated inside XML-delimited prompt boundaries (&lt;code&gt;&amp;lt;resume_document&amp;gt;&lt;/code&gt; and &lt;code&gt;&amp;lt;job_posting&amp;gt;&lt;/code&gt;), successfully neutralizing adversarial prompt injection attacks.&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
    </item>
    <item>
      <title>OutOfOffice AI: Step Away from Your Desk with Local Gemma 2 and Agent Telemetry</title>
      <dc:creator>Samar Shetye</dc:creator>
      <pubDate>Sun, 11 Oct 2026 15:28:18 +0000</pubDate>
      <link>https://dev.to/samar_shetye/outofoffice-ai-step-away-from-your-desk-with-local-gemma-2-and-agent-telemetry-2mam</link>
      <guid>https://dev.to/samar_shetye/outofoffice-ai-step-away-from-your-desk-with-local-gemma-2-and-agent-telemetry-2mam</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-week1-2026-10-05"&gt;Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What I Built
&lt;/h2&gt;

&lt;p&gt;Most modern AI coding assistants do the exact opposite of encouraging you to touch grass: they demand your constant attention. You write a prompt, watch tokens stream, babysit diffs line-by-line, and wait on progress spinners.&lt;br&gt;
&lt;strong&gt;OutOfOffice AI (OOO)&lt;/strong&gt; was engineered to break this screen dependency.&lt;br&gt;
It is an autonomous, local-first developer agent powered by &lt;strong&gt;Google Gemma 2&lt;/strong&gt;. Instead of micromanaging an AI in an IDE window, you:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Target any local Git repository and set a maintenance directive (e.g., prune dead code, fix broken imports, run test suites).&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Close your laptop, step away from your desk, and touch grass.&lt;/strong&gt;&lt;/li&gt;
&lt;li&gt;Return to an executive &lt;strong&gt;ElevenLabs voice debrief&lt;/strong&gt; summarizing codebase health and modified files, backed by verified Git patches isolated in dedicated branches.
By turning asynchronous coding chores over to an autonomous local agent, your screen time becomes the shortest part of the development loop.&lt;/li&gt;
&lt;/ol&gt;


&lt;h2&gt;
  
  
  Demo
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Youtube Video:&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;  &lt;iframe src="https://www.youtube.com/embed/AvlVI2Yrmj0" width="710" height="399"&gt;
  &lt;/iframe&gt;
&lt;/p&gt;

&lt;h3&gt;
  
  
  What You See in the Demo:
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;OOO Landing Page:&lt;/strong&gt; Clean, developer-focused overview showing the local-first architecture and guarantees.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Mission Control:&lt;/strong&gt; Local Git manifest intake and zero-data-egress model selection.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Sentry Telemetry Waterfall:&lt;/strong&gt; Microsecond span instrumentation across &lt;code&gt;PlannerNode&lt;/code&gt;, &lt;code&gt;ASTScanNode&lt;/code&gt;, and &lt;code&gt;GitBranchIsolation&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;ElevenLabs Voice Debrief:&lt;/strong&gt; Executive audio briefing summarizing repository health score and tasks completed.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Unified Diff Viewer:&lt;/strong&gt; Syntax-highlighted git diffs isolated in the &lt;code&gt;ooo/agent-audit&lt;/code&gt; branch.&lt;/li&gt;
&lt;/ul&gt;




&lt;h2&gt;
  
  
  Code
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;GitHub Repository:&lt;/strong&gt;&lt;br&gt;
&lt;/p&gt;
&lt;div class="ltag-github-readme-tag"&gt;
  &lt;div class="readme-overview"&gt;
    &lt;h2&gt;
      &lt;img src="https://assets.dev.to/assets/github-logo-5a155e1f9a670af7944dd5e12375bc76ed542ea80224905ecaf878b9157cdefc.svg" alt="GitHub logo"&gt;
      &lt;a href="https://github.com/Samar-365" rel="noopener noreferrer"&gt;
        Samar-365
      &lt;/a&gt; / &lt;a href="https://github.com/Samar-365/OutOfOffice-" rel="noopener noreferrer"&gt;
        OutOfOffice-
      &lt;/a&gt;
    &lt;/h2&gt;
    &lt;h3&gt;
      Local Autonomous Coding Agent
    &lt;/h3&gt;
  &lt;/div&gt;
  &lt;div class="ltag-github-body"&gt;
    
&lt;div id="readme" class="md"&gt;&lt;div class="markdown-heading"&gt;
&lt;h1 class="heading-element"&gt;OutOfOffice AI (OOO)&lt;/h1&gt;
&lt;/div&gt;
&lt;p&gt;Local-First Autonomous Coding Agent Powered by Google Gemma 2, Sentry Agent Tracing, and ElevenLabs.&lt;/p&gt;
&lt;p&gt;&lt;a rel="noopener noreferrer" href="https://github.com/Samar-365/OutOfOffice-/docs/banner.png"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fraw.githubusercontent.com%2FSamar-365%2FOutOfOffice-%2FHEAD%2Fdocs%2Fbanner.png" alt="OutOfOffice AI Banner"&gt;&lt;/a&gt;&lt;/p&gt;
&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;Overview&lt;/h2&gt;
&lt;/div&gt;
&lt;p&gt;Modern AI coding tools suffer from a common limitation: they demand constant developer presence. Developers spend hours waiting for tokens to stream, reviewing single-line completions, and babysitting chat prompts.&lt;/p&gt;
&lt;p&gt;OutOfOffice AI (OOO) flips this workflow. It is an autonomous, local-first developer agent designed for asynchronous execution:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;Target any local Git repository and set a maintenance directive (dead code pruning, AST refactoring, test repair).&lt;/li&gt;
&lt;li&gt;Step away from your desk while the agent executes locally on your hardware.&lt;/li&gt;
&lt;li&gt;Return to an executive ElevenLabs audio debrief and inspect verified, branch-isolated syntax diffs.&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;No proprietary code is transmitted to cloud LLM APIs. All model inference runs on your machine via Google Gemma 2.&lt;/p&gt;

&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;Core Guarantees&lt;/h2&gt;
&lt;/div&gt;
&lt;ul&gt;
&lt;li&gt;100% Local Intelligence: Runs Google Gemma 2 locally through Ollama. Zero cloud token costs, zero latency overhead, and complete data privacy.&lt;/li&gt;
&lt;li&gt;…&lt;/li&gt;
&lt;/ul&gt;&lt;/div&gt;
  &lt;/div&gt;
  &lt;div class="gh-btn-container"&gt;&lt;a class="gh-btn" href="https://github.com/Samar-365/OutOfOffice-" rel="noopener noreferrer"&gt;View on GitHub&lt;/a&gt;&lt;/div&gt;
&lt;/div&gt;


&lt;p&gt;The codebase is structured into:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;code&gt;backend/&lt;/code&gt;: FastAPI daemon, AST analyzer, Git manager, Sentry tracing setup, and Ollama client.&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;frontend/&lt;/code&gt;: React 18, TypeScript, Tailwind CSS, Vite, and Lucide icons.&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;fixtures/&lt;/code&gt;: Test repositories for validating dead-code pruning and AST syntax safety.&lt;/li&gt;
&lt;/ul&gt;




&lt;h2&gt;
  
  
  How I Built It
&lt;/h2&gt;

&lt;p&gt;OutOfOffice AI is built from the ground up around open-source AI and observability:&lt;/p&gt;

&lt;h3&gt;
  
  
  1. Local Intelligence with Google Gemma 2
&lt;/h3&gt;

&lt;p&gt;The core reasoning engine runs &lt;strong&gt;Google Gemma 2:2b&lt;/strong&gt; locally through Ollama (&lt;code&gt;http://localhost:11434&lt;/code&gt;). The agent parses the target repository's Abstract Syntax Tree (AST), identifies unused imports or unreferenced functions, and formulates multi-step execution plans completely offline.&lt;/p&gt;

&lt;h3&gt;
  
  
  2. Distributed Telemetry with Sentry Agent Tracing
&lt;/h3&gt;

&lt;p&gt;Autonomous agents shouldn't be black boxes. We instrumented the entire Python agent using Sentry's distributed tracing SDK. Every node in the execution graph is tracked with microsecond span precision:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;code&gt;PlannerNode&lt;/code&gt;: Latency and prompt token metrics.&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;ASTScanNode&lt;/code&gt;: File manifest and syntax structure inspection.&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;GitBranchIsolation&lt;/code&gt;: Creation of isolated branches (&lt;code&gt;ooo/agent-audit&lt;/code&gt;).&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;VerificationNode&lt;/code&gt;: AST validation preventing syntax errors before patch emission.
### 3. Hands-Free Voice Debrief with ElevenLabs
When returning to your desk, you don't need to dig through thousands of lines of terminal output. The backend synthesizes an executive voice summary using the ElevenLabs Turbo API (&lt;code&gt;eleven_turbo_v2_5&lt;/code&gt;), providing a hands-free report of codebase health, modified files, and test results.&lt;/li&gt;
&lt;/ul&gt;




&lt;h2&gt;
  
  
  Why Does Open Innovation Matter?
&lt;/h2&gt;

&lt;p&gt;Open innovation is essential to the core value of OutOfOffice AI:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;True Data Privacy:&lt;/strong&gt; Closed cloud APIs require uploading private source code and internal architectures to remote servers. With open-weight models like &lt;strong&gt;Google Gemma 2&lt;/strong&gt;, your proprietary code never leaves your local machine.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Zero Token Cost at Scale:&lt;/strong&gt; Autonomous coding loops perform multiple iterations—scanning dozens of files, verifying ASTs, and testing patches. On metered cloud APIs, this becomes cost-prohibitive. Open-weight models running on local hardware cost $0.00.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Offline Resilience:&lt;/strong&gt; Developers can work on planes, trains, or trails with zero internet connection. Open-source AI transforms intelligence from a metered subscription into durable, local computing infrastructure.&lt;/li&gt;
&lt;/ul&gt;




&lt;h2&gt;
  
  
  Prize Categories
&lt;/h2&gt;

&lt;p&gt;I am entering the following categories:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Best Use of Gemma ($200):&lt;/strong&gt; OutOfOffice AI uses Google's open-weight &lt;strong&gt;Gemma 2:2b&lt;/strong&gt; running locally via Ollama as the core intelligence engine for AST analysis and code refactoring.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Best Use of Sentry Agent Tracing ($100):&lt;/strong&gt; Full end-to-end distributed tracing across all agent reasoning nodes and tools, with microsecond span waterfalls and live telemetry.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Best Use of ElevenLabs ($100):&lt;/strong&gt; Synthesizes an executive hands-free audio debrief of codebase health and completed tasks, eliminating the need to read terminal logs when returning to the desk.&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
    </item>
    <item>
      <title>CodeMap AI — Helping My Friend Understand Unfamiliar Codebases</title>
      <dc:creator>Samar Shetye</dc:creator>
      <pubDate>Sat, 03 Oct 2026 18:11:25 +0000</pubDate>
      <link>https://dev.to/samar_shetye/codemap-ai-helping-my-friend-understand-unfamiliar-codebases-177d</link>
      <guid>https://dev.to/samar_shetye/codemap-ai-helping-my-friend-understand-unfamiliar-codebases-177d</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-weekend-2026-10-01"&gt;Hacktoberfest Weekend Challenge: Build for a Friend&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What I Built
&lt;/h2&gt;

&lt;p&gt;When developers join an unfamiliar project, one of the first challenges is understanding how everything fits together.&lt;br&gt;
You open the repository and see something like:&lt;br&gt;
&lt;code&gt;frontend/&lt;br&gt;
backend/&lt;br&gt;
components/&lt;br&gt;
services/&lt;br&gt;
routes/&lt;br&gt;
controllers/&lt;br&gt;
database/&lt;br&gt;
utils/&lt;br&gt;
&lt;/code&gt;&lt;br&gt;
Then the real questions begin:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Where does the application start?&lt;/li&gt;
&lt;li&gt;Which component calls this API?&lt;/li&gt;
&lt;li&gt;How does data flow from the frontend to the database?&lt;/li&gt;
&lt;li&gt;What files are responsible for authentication?&lt;/li&gt;
&lt;li&gt;If I change this file, what else could be affected?&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;I built CodeMap AI for a friend who was struggling with exactly this problem.&lt;br&gt;
Instead of making them manually explore an unfamiliar repository, CodeMap AI analyzes the project and turns its structure and relationships into an interactive visual codebase map.&lt;br&gt;
I had my friend try CodeMap AI on an unfamiliar project to see whether the visual map and AI explanations actually made the codebase easier to understand.&lt;br&gt;
Using the visual map made it easier for him to understand how the different parts of the project are connected. Instead of opening files one by one, the overall structure could be understood first, followed by the AI explanation for specific parts of the code.&lt;/p&gt;

&lt;p&gt;How it works:&lt;br&gt;
The user provides a project/repository:&lt;br&gt;
GitHub Repository&lt;br&gt;
        ↓&lt;br&gt;
CodeMap AI&lt;br&gt;
        ↓&lt;br&gt;
Code Analysis&lt;br&gt;
        ↓&lt;br&gt;
Dependency &amp;amp; Relationship Detection&lt;br&gt;
        ↓&lt;br&gt;
Interactive Codebase Map&lt;br&gt;
        ↓&lt;br&gt;
AI-powered explanations&lt;/p&gt;

&lt;p&gt;The resulting map lets the user visually explore relationships between different parts of the application.&lt;/p&gt;
&lt;h2&gt;
  
  
  Demo
&lt;/h2&gt;

&lt;p&gt;Youtube video:&lt;/p&gt;

&lt;p&gt;  &lt;iframe src="https://www.youtube.com/embed/d1Heryv2LY0" width="710" height="399"&gt;
  &lt;/iframe&gt;
&lt;/p&gt;

&lt;h2&gt;
  
  
  Code
&lt;/h2&gt;

&lt;p&gt;GitHub Repository:&lt;/p&gt;


&lt;div class="ltag-github-readme-tag"&gt;
  &lt;div class="readme-overview"&gt;
    &lt;h2&gt;
      &lt;img src="https://assets.dev.to/assets/github-logo-5a155e1f9a670af7944dd5e12375bc76ed542ea80224905ecaf878b9157cdefc.svg" alt="GitHub logo"&gt;
      &lt;a href="https://github.com/Samar-365" rel="noopener noreferrer"&gt;
        Samar-365
      &lt;/a&gt; / &lt;a href="https://github.com/Samar-365/CodeMap" rel="noopener noreferrer"&gt;
        CodeMap
      &lt;/a&gt;
    &lt;/h2&gt;
    &lt;h3&gt;
      CodeMap AI is an AI-powered application that helps developers understand unfamiliar software projects by automatically analyzing a codebase and generating an interactive visual map of its architecture, components, dependencies, and data flow.
    &lt;/h3&gt;
  &lt;/div&gt;
  &lt;div class="ltag-github-body"&gt;
    
&lt;div id="readme" class="md"&gt;&lt;div class="markdown-heading"&gt;
&lt;h1 class="heading-element"&gt;CodeMap AI&lt;/h1&gt;
&lt;/div&gt;
&lt;p&gt;&lt;a href="https://hacktoberfest.com/" rel="nofollow noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/8ad61ad2998017b6c2ead0ca0903ddacc4da5b3be0dca8113d3847ab36108ebb/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f4861636b746f626572666573742d323032362532304275696c64253230666f7225323061253230467269656e642d6f72616e67652e737667" alt="Hacktoberfest 2026"&gt;&lt;/a&gt;
&lt;a href="https://github.com/Samar-365/CodeMap/LICENSE" rel="noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/08cef40a9105b6526ca22088bc514fbfdbc9aac1ddbf8d4e6c750e3a88a44dca/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f4c6963656e73652d4d49542d626c75652e737667" alt="License: MIT"&gt;&lt;/a&gt;
&lt;a href="https://www.python.org/" rel="nofollow noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/9833615dcd5c4175da2140bbe363cb42af4faaeffc0624453b6154e68a60d0aa/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f507974686f6e2d332e31312b2d626c75652e737667" alt="Python 3.11+"&gt;&lt;/a&gt;
&lt;a href="https://fastapi.tiangolo.com/" rel="nofollow noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/4045578ca71c45bb003b8dbb693341b821a7396b82c64c226a252011bf36dc50/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f466173744150492d302e3131302b2d3030393638382e737667" alt="FastAPI"&gt;&lt;/a&gt;
&lt;a href="https://react.dev/" rel="nofollow noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/6f04bb4f91778c6748097e7a9b2ccabdcaa4e36e85649567f7c05aef38d864d5/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f52656163742d31382b2d3631444146422e737667" alt="React 18+"&gt;&lt;/a&gt;
&lt;a href="https://ollama.ai/" rel="nofollow noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/dde28c879a6ec9d73d9c0c0394c0534e3ac3dea3599a93d66a7cbf386935373e/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f4f6c6c616d612d4f70656e2d2d576569676874732d707572706c652e737667" alt="Ollama"&gt;&lt;/a&gt;&lt;/p&gt;
&lt;blockquote&gt;
&lt;p&gt;An AI-powered visual codebase navigator for understanding unfamiliar projects.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;Problem&lt;/h2&gt;
&lt;/div&gt;
&lt;p&gt;Joining a new engineering team, jumping into an open-source project, or inheriting legacy code often feels overwhelming:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Spaghetti Dependencies&lt;/strong&gt;: Developers spend hours or days manually following imports, function definitions, and API calls across dozens of folders.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Outdated Documentation&lt;/strong&gt;: Architecture diagrams in READMEs or wikis are frequently stale, incomplete, or missing entirely.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Fear of Breaking Changes&lt;/strong&gt;: Modifying a utility, route, or model is risky when you cannot clearly visualize all downstream consumers.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Privacy and IP Concerns&lt;/strong&gt;: Sending proprietary codebases to third-party cloud LLMs violates corporate NDAs, security compliance policies, and incurs recurring API costs.&lt;/li&gt;
&lt;/ul&gt;

&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;Solution&lt;/h2&gt;
&lt;/div&gt;
&lt;p&gt;&lt;strong&gt;CodeMap AI&lt;/strong&gt; transforms complex, unfamiliar repositories into an &lt;strong&gt;interactive, multi-tier architectural map&lt;/strong&gt; and provides a &lt;strong&gt;context-grounded local AI assistant&lt;/strong&gt;:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Automatic Multi-Tier Layout&lt;/strong&gt;: Categorizes files into clean architectural layers (&lt;em&gt;Frontend UI&lt;/em&gt;, &lt;em&gt;API Endpoints&lt;/em&gt;, &lt;em&gt;Services&lt;/em&gt;, &lt;em&gt;Data&lt;/em&gt;…&lt;/li&gt;
&lt;/ul&gt;&lt;/div&gt;
  &lt;/div&gt;
  &lt;div class="gh-btn-container"&gt;&lt;a class="gh-btn" href="https://github.com/Samar-365/CodeMap" rel="noopener noreferrer"&gt;View on GitHub&lt;/a&gt;&lt;/div&gt;
&lt;/div&gt;


&lt;h2&gt;
  
  
  How I Built It
&lt;/h2&gt;

&lt;p&gt;CodeMap AI combines open-weight AI, local semantic search, and code analysis into one pipeline:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;&lt;p&gt;Local LLM via Ollama&lt;br&gt;
I use Ollama to run open-weight models entirely on the user's machine. The default model is Gemma 2 (2B) — small enough to run on a laptop CPU, but capable enough to produce clear architectural explanations.&lt;br&gt;
You can swap it for qwen3:4b, llama3.2, or phi4-mini depending on your hardware, all via an environment variable. No API key, no subscription, no cost.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Semantic Embeddings via SentenceTransformers&lt;br&gt;
Every code chunk is embedded using all-MiniLM-L6-v2 — a lightweight, locally-run embedding model from Hugging Face. These embeddings are indexed in FAISS for sub-millisecond vector search.&lt;br&gt;
When you ask a question, the RAG pipeline retrieves the most semantically relevant code chunks from your repository and injects them directly into the LLM's prompt context, making every answer grounded in your actual codebase.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Multi-language Code Analysis&lt;br&gt;
The backend performs multi-language code analysis across Python, JavaScript, TypeScript, JSX/TSX, Java, JSON, YAML, and SQL files using Python's native &lt;code&gt;ast&lt;/code&gt; module for Python code and custom tokenization/parsing logic for other supported file types. It extracts:&lt;br&gt;
Imports and dependencies&lt;br&gt;
Class and function definitions&lt;br&gt;
API route handlers (@app.get, Express routes, etc.)&lt;br&gt;
Frontend network calls (fetch, axios)&lt;br&gt;
Database schema definitions&lt;br&gt;
These are then resolved into a directed dependency graph that feeds both the React Flow visualization and the RAG retrieval pipeline.&lt;/p&gt;&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Tech Stack:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Frontend: React 18 + Vite + React Flow (Dagre)&lt;/li&gt;
&lt;li&gt;Backend:  Python 3.11 + FastAPI + Uvicorn&lt;/li&gt;
&lt;li&gt;AST Parsing:  Python ast module + regex tokenizers&lt;/li&gt;
&lt;li&gt;Vector Search:    FAISS + sentence-transformers&lt;/li&gt;
&lt;li&gt;Embeddings:   all-MiniLM-L6-v2 (local, free)&lt;/li&gt;
&lt;li&gt;Local LLM:    Ollama — qwen3:4b, gemma2:2b, llama3.2&lt;/li&gt;
&lt;li&gt;Graph Layout: Dagre (hierarchical DAG algorithm)&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Why Does Open Innovation Matter?
&lt;/h2&gt;

&lt;p&gt;One of the most important aspects of CodeMap AI is that the application deals with source code.&lt;br&gt;
A developer may want to understand a private project, a college project, or an internal codebase. Sending an entire repository to a closed AI service isn't always desirable.&lt;br&gt;
Using open-source/open-weight AI gives the project more flexibility around how the AI is run and where the source code is processed.&lt;br&gt;
For CodeMap AI, this means the system can be designed around:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Open-weight models&lt;/li&gt;
&lt;li&gt;Local inference&lt;/li&gt;
&lt;li&gt;Self-hosted AI&lt;/li&gt;
&lt;li&gt;Replaceable models&lt;/li&gt;
&lt;li&gt;Greater control over source-code processing&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The goal isn't simply to add AI to a developer tool.&lt;br&gt;
The goal is to make AI-based codebase understanding possible while giving developers more control over the models and infrastructure involved.&lt;br&gt;
In my implementation, the AI inference can run locally, meaning the source code doesn't have to be sent to a closed third-party AI provider for analysis.&lt;/p&gt;

&lt;h2&gt;
  
  
  Prize Categories
&lt;/h2&gt;

&lt;p&gt;Gemma — Best Use of Gemma&lt;br&gt;
I used Gemma, Google's open-weight model, as a core part of CodeMap AI's codebase understanding and AI-powered explanation system.&lt;br&gt;
ElevenLabs — Best Use of ElevenLabs&lt;br&gt;
I used ElevenLabs to generate the voice-over narration for my CodeMap AI demo video.&lt;br&gt;
The voice-over walks through the problem CodeMap AI solves, demonstrates the application's workflow, and explains how the visual codebase map and AI-powered analysis work.&lt;/p&gt;

</description>
      <category>devchallenge</category>
      <category>weekendchallenge</category>
      <category>hf26challenge</category>
    </item>
    <item>
      <title>Breaking the Language Barrier When Truth Matters Most</title>
      <dc:creator>Samar Shetye</dc:creator>
      <pubDate>Sun, 15 Mar 2026 20:37:37 +0000</pubDate>
      <link>https://dev.to/samar_shetye/breaking-the-language-barrier-when-truth-matters-most-5b5o</link>
      <guid>https://dev.to/samar_shetye/breaking-the-language-barrier-when-truth-matters-most-5b5o</guid>
      <description>&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2F3vflnuthwzot6o6tyz11.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2F3vflnuthwzot6o6tyz11.png" alt=" "&gt;&lt;/a&gt;When global conflicts ignite, the first casualty is often the truth. Local propaganda machines spin narratives, external news sites are blocked, and critical updates from international journalists remain trapped behind a massive barrier: language.&lt;/p&gt;

&lt;p&gt;Before I started coding LingualNews, my "blank canvas" was shaped by this massive information asymmetry. I needed to build a solution using a fast frontend (React/Vite) wrapped around a tight set of APIs (Lingo.dev and Groq) with one core mission: What if anyone, anywhere, could paste a URL from any global news outlet and instantly read—or listen to it—in their native tongue?&lt;/p&gt;

&lt;p&gt;The problem I am solving is access to unfiltered reality. LingualNews makes the world a better place by ensuring that truth is never paywalled by your geographic location or the language you happen to speak.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The High-Level Architecture&lt;/strong&gt;&lt;br&gt;
Think of LingualNews as a "Smart Proxy Funnel." You input a foreign news URL. The React client dynamically bypasses CORS constraints through fallback proxies. The raw chaotic HTML is cleaned by a heuristic DOM scraper, stripping away ads and noise. The isolated text is then pushed into a dual-engine AI pipeline: Lingo.dev handles the precise translation, while Groq (LLaMA 3.1) generates concise summaries and acts as a translation fallback. Finally, Native Web Speech APIs vocalize the translated text.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Where the "Magic" Happens&lt;/strong&gt;&lt;br&gt;
The real magic lives in scraperService.js and translationService.js. These modules transform chaotic, unstructured, ad-riddled foreign websites into clean, localized, readable components in milliseconds.&lt;/p&gt;

&lt;p&gt;If I had to isolate the biggest technical breakthrough in the app, it's just a few lines of code in the translation layer. During a geopolitical crisis, traffic spikes. APIs fail. If my primary translation API drops, my app cannot break.&lt;/p&gt;

&lt;p&gt;Here is my breakthrough, explained line-by-line:&lt;/p&gt;

&lt;p&gt;javascript&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="p"&gt;}&lt;/span&gt; &lt;span class="k"&gt;catch &lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;err&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;
      &lt;span class="nx"&gt;console&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;warn&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="dl"&gt;'&lt;/span&gt;&lt;span class="s1"&gt;Lingo.dev translation failed, falling back to Groq:&lt;/span&gt;&lt;span class="dl"&gt;'&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nx"&gt;err&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;message&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
      &lt;span class="c1"&gt;// Fallback to Groq&lt;/span&gt;
      &lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="nx"&gt;translateWithGroq&lt;/span&gt; &lt;span class="p"&gt;}&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="k"&gt;import&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="dl"&gt;'&lt;/span&gt;&lt;span class="s1"&gt;./aiService.js&lt;/span&gt;&lt;span class="dl"&gt;'&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
      &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="k"&gt;await&lt;/span&gt; 
      &lt;span class="nf"&gt;translateWithGroq&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;text&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nx"&gt;targetLang&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;span class="p"&gt;}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;ol&gt;
&lt;li&gt;catch (err): We intercept any failure from the primary Lingo.dev API.&lt;/li&gt;
&lt;li&gt;console.warn(...): Gracefully log the error without crashing the user's interface.&lt;/li&gt;
&lt;li&gt;const { translateWithGroq } = await import(...): This is the secret sauce. Instead of loading the heavy AI service upfront, we dynamically lazy-load the Groq module only if Lingo.dev fails. This keeps the initial bundle size incredibly small and fast.&lt;/li&gt;
&lt;li&gt;return await translateWithGroq(...): We leverage LLaMA 3.1 via Groq to step in as a secondary line of defense, ensuring the user instantly gets their localized news, no matter what.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;&lt;strong&gt;The Most Frustrating Bug&lt;/strong&gt;&lt;br&gt;
My biggest headache? Integrating the official Lingo.dev SDK. It was built for Node environments and threw obscure bundling errors that completely broke the Vite build. The undocumented, obscure fix? I threw away the official SDK entirely. I reverse-engineered their REST endpoint out of their docs and crafted a raw fetch POST request with a custom X-API-Key header. Sometimes, raw HTTP beats a polished SDK.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;When Clean Design Met Messy Reality&lt;/strong&gt;&lt;br&gt;
My initial, clean design assumed a simple querySelector('article') would cleanly extract text from any news site perfectly. The reality? Web scraping is a nightmare of custom ad-blockers, captchas, and chaotic div nests.&lt;/p&gt;

&lt;p&gt;My clean design broke immediately. I was forced to write a massive, messy array of heuristic CSS selectors and "noise" filters (.cookie-banner, .advertisement). When all else fails, I built a highly functional workaround: dumping the raw HTML directly into Groq AI with a strict prompt: "Return ONLY the article body text."&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Technical Debt &amp;amp; Trade-Offs&lt;/strong&gt;&lt;br&gt;
To cross the finish line quickly, I accepted a notable piece of technical debt: embedding API keys in the frontend environment (import.meta.env). While Vite obscures them in dev, client-side fetches are risky at scale. I accepted this to launch a fast MVP, knowing I can shift to edge functions in V2.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Try It Yourself&lt;/strong&gt;&lt;br&gt;
You can run this locally with practically zero friction:&lt;/p&gt;

&lt;p&gt;bash&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;git clone https://github.com/Samar-365/LingualNews
&lt;span class="nb"&gt;cd &lt;/span&gt;LingualNews
npm &lt;span class="nb"&gt;install
&lt;/span&gt;Add your .env with VITE_GROQ_API_KEY and VITE_LINGO_API_KEY
npm run dev
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;strong&gt;What’s Next?&lt;/strong&gt;&lt;br&gt;
If you fork this repository today, the most obvious missing piece to build is Offline PWA Support (Service Workers). Reading the news shouldn't require a constant, stable internet connection—especially for users in volatile regions. Local caching features could dramatically improve this app's life-saving potential.&lt;/p&gt;

&lt;p&gt;Code is more than syntax; it's how we connect the world. What are you building today?&lt;/p&gt;

&lt;p&gt;Check out the repo at &lt;a href="https://github.com/Samar-365/LingualNews" rel="noopener noreferrer"&gt;https://github.com/Samar-365/LingualNews&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Special thanks to &lt;a class="mentioned-user" href="https://dev.to/sumitsaurabh927"&gt;@sumitsaurabh927&lt;/a&gt; and &lt;a class="mentioned-user" href="https://dev.to/maxprilutskiy"&gt;@maxprilutskiy&lt;/a&gt; for their continuous guidance throughout the hackathon and also for providing us this great opportunity.&lt;/p&gt;

&lt;p&gt;Happy coding!&lt;/p&gt;

</description>
      <category>lingodotdev</category>
      <category>ai</category>
      <category>webdev</category>
      <category>javascript</category>
    </item>
    <item>
      <title>Stop Copy-Pasting from Images: Build a Universal Screen Translator with Python</title>
      <dc:creator>Samar Shetye</dc:creator>
      <pubDate>Thu, 05 Feb 2026 18:33:50 +0000</pubDate>
      <link>https://dev.to/samar_shetye/stop-copy-pasting-from-images-build-a-universal-screen-translator-with-python-54gm</link>
      <guid>https://dev.to/samar_shetye/stop-copy-pasting-from-images-build-a-universal-screen-translator-with-python-54gm</guid>
      <description>&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2F5wmxocgu79vtbsp75oh7.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2F5wmxocgu79vtbsp75oh7.png" alt=" " width="750" height="275"&gt;&lt;/a&gt;&lt;strong&gt;Lingo-Live&lt;/strong&gt; started with a frustration I’m sure you’ve felt too.&lt;/p&gt;

&lt;p&gt;Have you ever tried copying text from a YouTube video?&lt;br&gt;
Or translating a Japanese error message inside a game?&lt;/p&gt;

&lt;p&gt;Yeah. You can’t.&lt;br&gt;
Because it’s not text — it’s just pixels.&lt;/p&gt;

&lt;p&gt;Most of us end up doing one of two things:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;painfully typing everything by hand, or&lt;/li&gt;
&lt;li&gt;pulling out our phones and using Google Lens, holding it up to the screen like it’s 2010.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;It’s clunky. It breaks focus. And honestly, we can do better.&lt;/p&gt;

&lt;p&gt;So I built Lingo-Live — a sleek desktop app that lets you translate anything you see on your screen instantly.&lt;/p&gt;
&lt;h2&gt;
  
  
  The Superpower We Wanted
&lt;/h2&gt;

&lt;p&gt;I didn’t want just another translation app. I wanted something that felt like a superpower.&lt;/p&gt;
&lt;h2&gt;
  
  
  That meant it had to be:
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Invisible – runs quietly in the background&lt;/li&gt;
&lt;li&gt;Instant – hit a hotkey, select an area, get a translation&lt;/li&gt;
&lt;li&gt;Modern – glassy UI, dark mode, blur effects, no Windows-95 vibes&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Press Ctrl + Alt + T, drag over any part of your screen, and boom — translated text appears on top of whatever you’re doing.&lt;/p&gt;
&lt;h2&gt;
  
  
  The Secret Sauce: How Lingo-Live Works
&lt;/h2&gt;

&lt;p&gt;Python made this possible. It’s basically a Swiss Army knife for building tools like this.&lt;/p&gt;

&lt;p&gt;Here’s how everything comes together.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;1. The “Glass” Overlay&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;The trickiest part was creating a window that stays on top without being annoying.&lt;/p&gt;

&lt;p&gt;I used CustomTkinter to build a frameless, translucent overlay that feels light and modern.&lt;/p&gt;

&lt;p&gt;Key details:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Always on top so translations stay visible&lt;/li&gt;
&lt;li&gt;Semi-transparent so you can still see context underneath&lt;/li&gt;
&lt;li&gt;Frameless — no ugly title bar; custom drag-and-drop instead&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The result feels less like an app and more like a layer on your desktop.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;2. The Eyes (OCR)&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;When you trigger the hotkey, Lingo-Live doesn’t try to “read the screen.”&lt;/p&gt;

&lt;p&gt;Instead, it:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Lets you select a region&lt;/li&gt;
&lt;li&gt;Takes a screenshot of just that area&lt;/li&gt;
&lt;li&gt;Sends it to Tesseract OCR to extract text from the pixels&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Conceptually, it looks like this:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;screenshot = ImageGrab.grab(bbox=(x1, y1, x2, y2))
text = ocr_engine.extract_text(screenshot)
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;That’s where the magic starts — turning images into actual text.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;3. The Brain (Translation)&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Once OCR gives us something like こんにちは, we need a translation that actually makes sense.&lt;/p&gt;

&lt;p&gt;This is where &lt;strong&gt;Lingo.dev&lt;/strong&gt; comes in.&lt;/p&gt;

&lt;p&gt;Instead of raw dictionary swaps, it handles context properly, which makes a huge difference — especially for UI text, error messages, and game dialogue.&lt;/p&gt;

&lt;p&gt;The result feels natural, not robotic.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;4. The Voice (Text-to-Speech)&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Sometimes you don’t want to read. You just want to hear it.&lt;/p&gt;

&lt;p&gt;So I added Edge TTS, which uses the same high-quality voices found in Microsoft Edge.&lt;/p&gt;

&lt;p&gt;Now Lingo-Live can read translations out loud — great for pronunciation or just staying hands-free.&lt;/p&gt;

&lt;p&gt;“Fish are vertebrate animals that live in water…” &lt;/p&gt;

&lt;p&gt;&lt;strong&gt;5. Leveling Up: AI Summarization&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Full translations are great, but sometimes you just want the gist.&lt;/p&gt;

&lt;p&gt;So I added a Summarize button powered by Google Gemini.&lt;/p&gt;

&lt;p&gt;Here’s what happens:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;The translated text is sent to Gemini&lt;/li&gt;
&lt;li&gt;It returns a clean, one-sentence summary&lt;/li&gt;
&lt;li&gt;You get the point instantly
Perfect for skimming foreign articles, long error messages, or RPG dialogue dumps.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;strong&gt;6. Make It Yours: Settings That Actually Matter&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;I didn’t want Lingo-Live to feel rigid, so I built a full settings system backed by JSON.&lt;/p&gt;

&lt;p&gt;You can:&lt;/p&gt;

&lt;p&gt;[- Change the hotkey (Alt + Z? Sure.)&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Switch themes (dark mode is the correct choice)&lt;/li&gt;
&lt;li&gt;Pick different fonts (Roboto &amp;gt; Segoe UI, fight me)](url)&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Best part?&lt;br&gt;
All changes apply instantly — no restarts, no reloads. &lt;/p&gt;

&lt;h2&gt;
  
  
  Conclusion
&lt;/h2&gt;

&lt;p&gt;Building your own tools is one of the most satisfying parts of being a developer.&lt;/p&gt;

&lt;p&gt;Lingo-Live solves a problem I run into constantly: text that’s trapped inside images, videos, and games. Instead of working around it, I built something that feels fast, modern, and genuinely useful.&lt;/p&gt;

&lt;p&gt;If you’ve ever rage-typed a foreign error message at 2 AM, this app is for you.&lt;/p&gt;

&lt;p&gt;Lingo.dev makes localization feel effortless—turning a painful, error-prone task into a smooth, developer-friendly experience.&lt;/p&gt;

&lt;p&gt;Check out the repo at &lt;a href="https://github.com/Samar-365/lingo_live" rel="noopener noreferrer"&gt;https://github.com/Samar-365/lingo_live&lt;/a&gt;, clone the code, and stop copy-pasting from pixels.&lt;/p&gt;

&lt;p&gt;Special thanks to &lt;a class="mentioned-user" href="https://dev.to/sumitsaurabh927"&gt;@sumitsaurabh927&lt;/a&gt; and &lt;a class="mentioned-user" href="https://dev.to/maxprilutskiy"&gt;@maxprilutskiy&lt;/a&gt; for their continuous guidance throughout the hackathon and also for providing us this great opportunity.&lt;/p&gt;

&lt;p&gt;Happy coding! &lt;/p&gt;

</description>
      <category>programming</category>
      <category>webdev</category>
      <category>python</category>
      <category>lingodotdev</category>
    </item>
  </channel>
</rss>
