<?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: Mayur P</title>
    <description>The latest articles on DEV Community by Mayur P (@mayurponly).</description>
    <link>https://dev.to/mayurponly</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%2F4155252%2F56fb4e84-e440-4000-b06e-592387367861.jpg</url>
      <title>DEV Community: Mayur P</title>
      <link>https://dev.to/mayurponly</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/mayurponly"/>
    <language>en</language>
    <item>
      <title>Echoes AI: A Private, Local-First Memory Journal &amp; Story Sanctuary</title>
      <dc:creator>Mayur P</dc:creator>
      <pubDate>Wed, 28 Oct 2026 02:52:47 +0000</pubDate>
      <link>https://dev.to/mayurponly/echoes-ai-a-private-local-first-memory-journal-story-sanctuary-102b</link>
      <guid>https://dev.to/mayurponly/echoes-ai-a-private-local-first-memory-journal-story-sanctuary-102b</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 family has stories that slowly drift toward the edge of being forgotten: the quiet joke shared on a front porch during a summer storm, the secret twist to a holiday recipe, or the reassuring words of a grandparent whose voice was never recorded.&lt;/p&gt;

&lt;p&gt;For the &lt;strong&gt;Week 4 “Ghost in the Machine”&lt;/strong&gt; challenge, I built &lt;strong&gt;Echoes AI&lt;/strong&gt; — an open-source, private, local-first digital memory journal and story preserver. &lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Tagline:&lt;/strong&gt; &lt;em&gt;“Every memory deserves to live on.”&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;Instead of interpreting "Ghost in the Machine" through dark sci-fi or eerie necromancy (trying to create "AI avatars" that impersonate the deceased), Echoes AI approaches the theme with deep reverence and warmth. The "ghosts" in our machines shouldn't be synthetic copies pretending to speak for lost loved ones; they should be the living, cherished echoes of human memories, traditions, and life stories, preserved faithfully under the storyteller's absolute ownership.&lt;/p&gt;

&lt;h3&gt;
  
  
  Core Capabilities:
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;📝 Memory Editor &amp;amp; Fidelity Preservation:&lt;/strong&gt; Input rough notes, approximate dates, family relationships, categories, and tags. Preserves the user's authentic, unedited raw notes alongside any AI wording.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;✨ AI Story Enhancement:&lt;/strong&gt; Transforms rough notes into full, touching narratives across three selectable voices: &lt;em&gt;Warm &amp;amp; Heartfelt&lt;/em&gt;, &lt;em&gt;Simple &amp;amp; Conversational&lt;/em&gt;, and &lt;em&gt;Nostalgic &amp;amp; Reflective&lt;/em&gt;. Strictly preserves facts without hallucinating unmentioned events or tragedies.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;💭 Gentle Oral History Interview Assistant:&lt;/strong&gt; Formulates 5 gentle, non-intrusive questions (e.g., sensory details, family traditions, exact quotes) that users can answer to organically expand their memories.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;📚 SQLite Memory Library:&lt;/strong&gt; Full-text search and filtering by category or tag, with dual side-by-side inspection of original notes and the enhanced narrative.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;🕯️ Remembrance Sanctuary:&lt;/strong&gt; Creates dignified memorial pages honoring loved ones with life dates, favorite sayings, important traditions, and linked stories.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;📦 Verified Export &amp;amp; Restores:&lt;/strong&gt; One-click exports to Markdown (&lt;code&gt;.md&lt;/code&gt;) or text (&lt;code&gt;.txt&lt;/code&gt;), plus validated JSON full-database backups with duplicate protection.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;🛡️ 100% Privacy &amp;amp; Local Sovereignty:&lt;/strong&gt; Zero cloud API calls, zero telemetry, zero model training on user grief, and complete confirmation-guarded data deletion.&lt;/li&gt;
&lt;/ul&gt;




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

&lt;p&gt;Here is a look at &lt;strong&gt;Echoes AI&lt;/strong&gt; running locally in Streamlit with its warm scrapbook aesthetic:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;+-----------------------------------------------------------------------------------------+
|  📖 Echoes AI                                                                           |
|  “Every memory deserves to live on.”                                                    |
|  A Private, Local-First Memory Journal &amp;amp; Story Preserver                                |
+-----------------------------------------------------------------------------------------+
| [Navigation]                                                                            |
|  (*) Write &amp;amp; Enhance Memory   ( ) Memory Library   ( ) Remembrance Page   ( ) Backup    |
|                                                                                         |
|  Memory Title: Sunday Lemonade on the Screened Porch                                    |
|  Category: Childhood | Date: Summer 1994 | People: Grandma Clara, Cousin Leo            |
|                                                                                         |
|  Raw Notes:                                                                             |
|  &amp;gt; Hot July afternoon in Ohio. Cicadas loud in maple trees. Grandma squeezed lemons     |
|  &amp;gt; into cracked yellow pitcher. Used cane sugar &amp;amp; garden mint. Creaky cedar swing.      |
|                                                                                         |
|  [✨ AI Narrative Preview - Warm &amp;amp; Heartfelt Voice]                                     |
|  "The heavy heat of an Ohio July always seemed to soften the moment we stepped onto     |
|   Grandma Clara's screened porch. While the cicadas hummed their rhythmic summer song   |
|   high in the maple trees, Grandma was already at work with her beloved yellow ceramic  |
|   pitcher—its handle bearing familiar small chips from decades of Sunday gatherings.   |
|   With patient hands, she stirred fresh-squeezed lemons with sweet cane sugar and crisp |
|   spearmint plucked right from behind the shed..."                                      |
+-----------------------------------------------------------------------------------------+
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;


&lt;p&gt;You can run it locally in under 2 minutes:&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/song-code-won/echoes-ai.git
&lt;span class="nb"&gt;cd &lt;/span&gt;echoes-ai
pip &lt;span class="nb"&gt;install&lt;/span&gt; &lt;span class="nt"&gt;-r&lt;/span&gt; requirements.txt
streamlit run app.py
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&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/song-code-won" rel="noopener noreferrer"&gt;
        song-code-won
      &lt;/a&gt; / &lt;a href="https://github.com/song-code-won/echoes-ai" rel="noopener noreferrer"&gt;
        echoes-ai
      &lt;/a&gt;
    &lt;/h2&gt;
    &lt;h3&gt;
      Echoes AI — A Private, Local-First Memory Journal built for Hacktoberfest 2026 Ghost in the Machine
    &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;📖 Echoes AI — A Private, Local-First Memory Journal&lt;/h1&gt;
&lt;/div&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;em&gt;“Every memory deserves to live on.”&lt;/em&gt;&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;&lt;a href="https://github.com/song-code-won/echoes-ai/LICENSE" rel="noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/0a500a04a81b90bff4a4e668e58b20a718aafd06f89acb32c177eb003ed0daa3/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f4c6963656e73652d4d49542d62757267756e64792e737667" 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/1e84b5e6b27224195cfafe6258acef2a29506fb27c574ab58d217cb74de10851/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f507974686f6e2d332e31302532422d626c75652e737667" alt="Python: 3.10+"&gt;&lt;/a&gt;
&lt;a href="https://streamlit.io/" rel="nofollow noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/e8f2dcd1ce8eefeed5883f86b3b975e814601d3ed6039b8e8fefdabeb119c681/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f53747265616d6c69742d312e33352532422d7265642e737667" alt="Streamlit: 1.35+"&gt;&lt;/a&gt;
&lt;a href="https://ollama.com/" rel="nofollow noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/aab0a33393b836ed4794cadfee8975641f1dc25f420d053230449ef33b7e1183/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f4f6c6c616d612d4c6f63616c253230496e666572656e63652d6f72616e67652e737667" alt="Ollama: Local"&gt;&lt;/a&gt;
&lt;a href="https://github.com/song-code-won/echoes-ai/tests/" rel="noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/aefc96976fd8c908a1981e8692422db5d0b35a24df1af164b8f13290d302c3eb/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f54657374732d32302532307061737365642d677265656e2e737667" alt="Tests: 20 passed"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;🌟 Project Motivation&lt;/h2&gt;
&lt;/div&gt;
&lt;p&gt;Family stories, childhood traditions, and the quiet wisdom of loved ones are fragile. Too often, they slip away because we only have scattered notes, half-remembered fragments, or lack the time to craft them into a lasting narrative.&lt;/p&gt;
&lt;p&gt;Yet handing personal, emotionally intimate memories over to cloud-based AI services comes at an unacceptable privacy cost: personal grief and family history sent over third-party networks, analyzed, and ingested into corporate training models. Furthermore, many commercial tools cross ethical lines by promising creepy digital avatars or claiming to "talk to the dead."&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Echoes AI&lt;/strong&gt; was created for the &lt;strong&gt;Hacktoberfest 2026 DEV Open-Source AI Challenge, Week 4: “Ghost in the Machine”&lt;/strong&gt;. It is a respectful, quiet digital sanctuary that preserves true personal stories entirely on your own computer.&lt;/p&gt;

&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;🏛️ System Architecture&lt;/h2&gt;
&lt;/div&gt;

  &lt;div class="js-render-enrichment-target"&gt;
    &lt;div class="render-plaintext-hidden"&gt;
      &lt;pre&gt;flowchart TD
    User([User /&lt;/pre&gt;…&lt;/div&gt;
&lt;/div&gt;&lt;/div&gt;
  &lt;/div&gt;
  &lt;div class="gh-btn-container"&gt;&lt;a class="gh-btn" href="https://github.com/song-code-won/echoes-ai" rel="noopener noreferrer"&gt;View on GitHub&lt;/a&gt;&lt;/div&gt;
&lt;/div&gt;


&lt;p&gt;&lt;strong&gt;Repository:&lt;/strong&gt; &lt;a href="https://github.com/song-code-won/echoes-ai" rel="noopener noreferrer"&gt;https://github.com/song-code-won/echoes-ai&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  Key Project Architecture:
&lt;/h3&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;echoes-ai/
├── app.py                         # Streamlit UI with Cream &amp;amp; Burgundy Scrapbook Design
├── config.py                      # Local paths, model choices, narrative styles
├── services/
│   ├── memory_repository.py       # Thread-safe SQLite database manager
│   ├── ollama_client.py           # Local REST client (127.0.0.1:11434)
│   ├── story_writer.py            # Factual-fidelity narrative enhancer
│   ├── interview_questions.py     # Gentle oral history prompt generator
│   ├── remembrance_builder.py     # Dignified tribute page synthesizer
│   └── export_service.py          # Markdown, TXT &amp;amp; JSON backup/restore
├── prompts/                       # Transparent prompt templates
└── tests/                         # 20 automated pytest test cases (100% pass)
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;






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

&lt;p&gt;Echoes AI was engineered from the ground up around &lt;strong&gt;local, open-source AI&lt;/strong&gt; and &lt;strong&gt;zero-cloud architecture&lt;/strong&gt;:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Local LLM Engine (Ollama + Qwen 2.5):&lt;/strong&gt;

&lt;ul&gt;
&lt;li&gt;Inference runs via &lt;strong&gt;Ollama&lt;/strong&gt; on &lt;code&gt;http://127.0.0.1:11434&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;The default model is &lt;strong&gt;&lt;code&gt;qwen2.5:3b&lt;/code&gt;&lt;/strong&gt; (Apache 2.0 license), which provides incredible storytelling nuance, emotional warmth, and fact-following capability while requiring only ~3.5 GB of RAM/VRAM.&lt;/li&gt;
&lt;li&gt;Also supports &lt;code&gt;qwen2.5:7b&lt;/code&gt;, &lt;code&gt;llama3.2:3b&lt;/code&gt;, or &lt;code&gt;mistral:7b&lt;/code&gt; seamlessly through an in-app selector.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Deterministic Offline Fallbacks:&lt;/strong&gt;

&lt;ul&gt;
&lt;li&gt;If Ollama is offline or not installed, the app doesn't crash: it gracefully switches to curated offline interview questions and structured Markdown memorial templates, allowing core journaling to work 100% offline.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Streamlit Frontend with Custom Aesthetic CSS:&lt;/strong&gt;

&lt;ul&gt;
&lt;li&gt;Designed with warm cream (&lt;code&gt;#FAF7F2&lt;/code&gt;), muted burgundy (&lt;code&gt;#6B1D2F&lt;/code&gt;), and subtle marigold (&lt;code&gt;#D9822B&lt;/code&gt;) tones paired with serif typography (&lt;code&gt;Crimson Pro&lt;/code&gt;), creating a calm, respectful journal experience rather than a cold tech dashboard.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;SQLite &amp;amp; Data Integrity Layer:&lt;/strong&gt;

&lt;ul&gt;
&lt;li&gt;Custom thread-safe &lt;code&gt;MemoryRepository&lt;/code&gt; using native Python &lt;code&gt;sqlite3&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;Strict duplicate detection and payload validation when restoring from JSON backups.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Automated Testing Suite:&lt;/strong&gt;

&lt;ul&gt;
&lt;li&gt;20 comprehensive &lt;code&gt;pytest&lt;/code&gt; tests mocking AI responses to verify CRUD operations, search filters, prompt formatting, and error handling without requiring live model downloads.&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;When preserving personal stories, family traditions, and tributes to lost loved ones, &lt;strong&gt;open innovation and local models are not merely a nice feature — they are an ethical necessity&lt;/strong&gt;:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Freedom from Data Harvesting:&lt;/strong&gt; Cloud AI services frequently retain prompts or reserve rights to train future foundation models on user text. Personal grief, family disputes, and intimate childhood memories should &lt;strong&gt;never&lt;/strong&gt; become training weights for a tech conglomerate.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Immunity from Service Deprecation:&lt;/strong&gt; When you rely on a proprietary cloud API, your software breaks the day the provider shuts down an endpoint or changes billing tiers. An open-weight model like Qwen 2.5 running on your own CPU/GPU via Ollama will run 20 years from now exactly as it does today.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Ethical Guardrails You Control:&lt;/strong&gt; Closed commercial platforms often push "AI companions" that try to mimic deceased people. Open-source code allows us to establish firm, transparent boundaries: our prompt templates and architecture strictly prohibit digital impersonation while respecting genuine human memory.&lt;/li&gt;
&lt;/ol&gt;




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

&lt;p&gt;During testing, I entered raw notes about a family member's Sunday lemonade ritual — the yellow ceramic pitcher with chips on the handle, the sound of cicadas in the trees, and the quiet advice that &lt;em&gt;"patience makes sweet lemonade, but hurried lemonade always stays sour."&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;Seeing those scattered bullet points woven into a cohesive, warm story without inventing fake details or adding cheesy embellishments was genuinely moving. It captured the exact feeling of sitting on that wooden porch thirty years ago.&lt;/p&gt;

&lt;p&gt;That is the true spirit of &lt;strong&gt;Echoes AI&lt;/strong&gt;: the machine doesn't replace the human heart — it simply helps us hold on to the light our loved ones left behind.&lt;/p&gt;




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

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Primary Challenge:&lt;/strong&gt; Hacktoberfest Open-Source AI Challenge Week 4: “Ghost in the Machine”&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;General Track:&lt;/strong&gt; Open-Source AI Applications / Local-First AI&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
    </item>
    <item>
      <title>RepoSloth: Let Local AI Explore Codebases While You Take It Slow 🦥</title>
      <dc:creator>Mayur P</dc:creator>
      <pubDate>Sat, 24 Oct 2026 16:28:56 +0000</pubDate>
      <link>https://dev.to/mayurponly/reposloth-let-local-ai-explore-codebases-while-you-take-it-slow-c64</link>
      <guid>https://dev.to/mayurponly/reposloth-let-local-ai-explore-codebases-while-you-take-it-slow-c64</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;When landing on a new open-source repository or legacy codebase, developers typically engage in what I call the &lt;strong&gt;"Codebase Shuffle"&lt;/strong&gt;: opening dozens of folders, hunting for where &lt;code&gt;main&lt;/code&gt; or the router lives, guessing how database pools and authentication are configured, and reading thousands of lines of code just to build a baseline mental model.&lt;/p&gt;

&lt;p&gt;I built &lt;strong&gt;RepoSloth&lt;/strong&gt; to let developers be intentionally lazy and take it slow.&lt;/p&gt;

&lt;p&gt;RepoSloth is a private, local-first developer tool that ingests any local Git repository or uploaded ZIP archive and analyzes the codebase using open-weight AI models running on your own machine. &lt;/p&gt;

&lt;h3&gt;
  
  
  What RepoSloth does for you:
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Zero-Execution Codebase Synthesis&lt;/strong&gt;: Instantly categorizes frameworks, candidate entry points, and manifest dependencies, clearly separating observed facts from AI inferences.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;File &amp;amp; Function Explainer&lt;/strong&gt;: Breaks down complex modules into plain-English summaries, inputs/outputs, implementation flows, and potential edge cases.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Grounded "Ask Your Codebase" Q&amp;amp;A&lt;/strong&gt;: Uses localized windowed chunking and lexical retrieval to answer questions like &lt;em&gt;"Where is authentication handled?"&lt;/em&gt; or &lt;em&gt;"Where should I add a new endpoint?"&lt;/em&gt;, complete with verifiable file paths and line citations.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Automated Mermaid Architecture Diagrams&lt;/strong&gt;: Generates visual module diagrams and dependency maps strictly from verified import evidence.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Secret &amp;amp; Traversal Defense Shield&lt;/strong&gt;: Prevents Zip-Slip exploits on uploaded archives and redacts API keys, AWS credentials, Bearer tokens, and &lt;code&gt;.env&lt;/code&gt; files before code reaches the AI prompt.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Developers, open-source contributors, code reviewers, and engineering managers get to be lazier: RepoSloth does the tedious exploration while you sit back with a cup of coffee.&lt;/p&gt;




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

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;GitHub Repository&lt;/strong&gt;: &lt;a href="https://github.com/song-code-won/RepoSloth" rel="noopener noreferrer"&gt;https://github.com/song-code-won/RepoSloth&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Live Local Dashboard&lt;/strong&gt;: Run &lt;code&gt;streamlit run reposloth/app.py&lt;/code&gt; after pulling your local model via &lt;code&gt;ollama run qwen2.5-coder:7b&lt;/code&gt;.&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  Quick Walkthrough:
&lt;/h3&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Connect Local Engine&lt;/strong&gt;: RepoSloth automatically discovers your local Ollama daemon on &lt;code&gt;localhost:11434&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Scan Repository&lt;/strong&gt;: Select a local repository or drop in an archive (a sample FastAPI microservice is included in &lt;code&gt;/sample_repo&lt;/code&gt;).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Explore&lt;/strong&gt;:

&lt;ul&gt;
&lt;li&gt;The &lt;strong&gt;Overview&lt;/strong&gt; tab reveals the detected framework (FastAPI), entry point (&lt;code&gt;app.py&lt;/code&gt;), and dependencies.&lt;/li&gt;
&lt;li&gt;The &lt;strong&gt;Ask Your Codebase&lt;/strong&gt; tab retrieves &lt;code&gt;auth.py&lt;/code&gt; lines 1–17 to explain token verification with exact line citations.&lt;/li&gt;
&lt;li&gt;The &lt;strong&gt;Architecture&lt;/strong&gt; tab visualizes module relationships in an exportable Mermaid diagram.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;/ol&gt;




&lt;h2&gt;
  
  
  Code
&lt;/h2&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/song-code-won" rel="noopener noreferrer"&gt;
        song-code-won
      &lt;/a&gt; / &lt;a href="https://github.com/song-code-won/RepoSloth" rel="noopener noreferrer"&gt;
        RepoSloth
      &lt;/a&gt;
    &lt;/h2&gt;
    &lt;h3&gt;
      RepoSloth: Local AI Codebase Explainer using open-weight models (Qwen 2.5 Coder / Ollama). Built for Hacktoberfest 2026 DEV Challenge.
    &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;🦥 RepoSloth — Local AI Codebase Explainer&lt;/h1&gt;
&lt;/div&gt;
&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;"Let AI explore the code while you take it slow."&lt;/strong&gt;&lt;/p&gt;
&lt;/blockquote&gt;
&lt;p&gt;Built for the &lt;strong&gt;Hacktoberfest 2026 DEV Open-Source AI Challenge, Week 3: “Slow and Steady.”&lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;RepoSloth is a private, local-first developer tool that helps you instantly understand unfamiliar repositories without opening every single file. Powered by &lt;strong&gt;open-weight models&lt;/strong&gt; running locally through &lt;strong&gt;Ollama&lt;/strong&gt;, RepoSloth synthesizes architecture, maps entry points, explains complex functions, and answers codebase questions with verifiable line citations.&lt;/p&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;Zero-Execution Sandbox&lt;/strong&gt;: Never runs untrusted code, build scripts, or repository hooks.&lt;/li&gt;
&lt;li&gt;🔒 &lt;strong&gt;100% Local &amp;amp; Air-Gapped&lt;/strong&gt;: Runs entirely on your machine via Ollama. No closed API keys, subscription fees, or data leaks.&lt;/li&gt;
&lt;li&gt;🔐 &lt;strong&gt;Built-in Secret &amp;amp; Traversal Shield&lt;/strong&gt;: Redacts API keys, credentials, and &lt;code&gt;.env&lt;/code&gt; files automatically. Thwarts Zip-Slip path traversal attacks.&lt;/li&gt;
&lt;li&gt;📋 &lt;strong&gt;Automated Repository Overview&lt;/strong&gt;: Distinguishes verified facts (manifests, files, imports) from AI-generated hypotheses.&lt;/li&gt;
&lt;li&gt;🔍 &lt;strong&gt;File&lt;/strong&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/song-code-won/RepoSloth" rel="noopener noreferrer"&gt;View on GitHub&lt;/a&gt;&lt;/div&gt;
&lt;/div&gt;


&lt;p&gt;All code is open-source under the permissive &lt;strong&gt;MIT License&lt;/strong&gt;.&lt;/p&gt;




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

&lt;p&gt;RepoSloth is built using a local-first stack that prioritizes security, reproducibility, and zero external costs:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Local Inference Engine&lt;/strong&gt;: &lt;a href="https://ollama.com" rel="noopener noreferrer"&gt;Ollama&lt;/a&gt; running open-weight code models.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Primary AI Model&lt;/strong&gt;: &lt;strong&gt;Qwen 2.5 Coder (7B)&lt;/strong&gt; (Apache 2.0 license) — provides deep architectural comprehension and reasoning. For low-spec or CPU-only setups, &lt;strong&gt;Qwen 2.5 Coder (1.5B)&lt;/strong&gt; is supported out-of-the-box.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Frontend / Dashboard&lt;/strong&gt;: &lt;a href="https://streamlit.io/" rel="noopener noreferrer"&gt;Streamlit&lt;/a&gt; with custom developer styling, collapsible inspectors, and reactive tabs.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;File Discovery &amp;amp; Parsing&lt;/strong&gt;: Python standard library (&lt;code&gt;pathlib&lt;/code&gt;, &lt;code&gt;ast&lt;/code&gt;, &lt;code&gt;os&lt;/code&gt;) with intelligent directory exclusion (&lt;code&gt;node_modules&lt;/code&gt;, &lt;code&gt;venv&lt;/code&gt;, &lt;code&gt;.git&lt;/code&gt;, build outputs, and binaries).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Security &amp;amp; Secret Redactor&lt;/strong&gt;: A multi-pattern regex sanitizer that strips high-entropy credentials, private keys, and environment tokens.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Grounded Context Chunker&lt;/strong&gt;: A windowed source chunker that calculates lexical and keyword overlap to retrieve supporting snippets without requiring a heavyweight vector database.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Testing&lt;/strong&gt;: Automated &lt;code&gt;pytest&lt;/code&gt; suite with 14 passing tests verifying archive safety, secret redaction, and offline daemon error handling.&lt;/li&gt;
&lt;/ul&gt;




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

&lt;p&gt;For a codebase exploration tool, &lt;strong&gt;open innovation and local models are non-negotiable&lt;/strong&gt;:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Intellectual Property &amp;amp; Privacy&lt;/strong&gt;: Proprietary enterprise code, pre-release startup repos, and sensitive business logic cannot simply be piped to closed third-party cloud APIs. Open-weight models running on Ollama ensure that &lt;strong&gt;zero bytes of code ever leave the local machine&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Air-Gapped &amp;amp; Offline Reliability&lt;/strong&gt;: Once the open-weight model weights are downloaded, RepoSloth works completely offline—on flights, trains, or air-gapped corporate intranets.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;No Metered API Costs&lt;/strong&gt;: Analyzing a 50,000-line repository with closed APIs can consume dozens of dollars in token fees per run. With open-weight models, exploration is truly free and limitless.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Reproducibility &amp;amp; Longevity&lt;/strong&gt;: Closed APIs deprecate models and alter outputs without warning. Open-weight models with permissive licenses (like Apache 2.0) ensure that software built today remains functional tomorrow.&lt;/li&gt;
&lt;/ol&gt;




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

&lt;p&gt;Onboarding onto an unfamiliar 10,000+ line repository typically consumes &lt;strong&gt;2 to 4 hours&lt;/strong&gt; of manual file reading, tracing imports, and locating entry points before making a pull request.&lt;/p&gt;

&lt;p&gt;With RepoSloth:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Scan &amp;amp; Structure&lt;/strong&gt;: Completed in &lt;strong&gt;under 2 seconds&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Architecture &amp;amp; Entrypoint Detection&lt;/strong&gt;: Instant.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Targeted Code Questioning&lt;/strong&gt;: Answers with direct line numbers in &lt;strong&gt;under 15 seconds&lt;/strong&gt;.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;That cuts preliminary repository reconnaissance down from half a day to less than 2 minutes—saving hours of cognitive fatigue and letting you stay slow and steady!&lt;/p&gt;




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

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Hacktoberfest Week 3: Slow and Steady&lt;/strong&gt; (Primary Challenge Category)&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
    </item>
    <item>
      <title>PrioritiAI — Transforming Task Paralysis into Action with Open-Source Local AI</title>
      <dc:creator>Mayur P</dc:creator>
      <pubDate>Tue, 13 Oct 2026 02:49:55 +0000</pubDate>
      <link>https://dev.to/mayurponly/prioritiai-transforming-task-paralysis-into-action-with-open-source-local-ai-5eec</link>
      <guid>https://dev.to/mayurponly/prioritiai-transforming-task-paralysis-into-action-with-open-source-local-ai-5eec</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;PrioritiAI&lt;/strong&gt; is a privacy-first, open-source AI task management and scheduling web application that cuts through decision paralysis for students, freelancers, and small business owners.&lt;/p&gt;

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

&lt;p&gt;Traditional todo apps function as static checklists: they collect tasks, but leave users overwhelmed by the hardest question: &lt;em&gt;"What should I work on right now, and why?"&lt;/em&gt; When users face 15+ competing responsibilities, they experience decision fatigue and fall into procrastination or default to low-impact tasks. Meanwhile, closed cloud AI tools demand that users stream sensitive client notes, deadlines, and project details to third-party servers.&lt;/p&gt;

&lt;h3&gt;
  
  
  The Solution
&lt;/h3&gt;

&lt;p&gt;PrioritiAI introduces a personal AI productivity assistant that:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Identifies "Your Next Best Task"&lt;/strong&gt;: Evaluates tasks across &lt;strong&gt;Urgency&lt;/strong&gt;, &lt;strong&gt;Importance&lt;/strong&gt;, &lt;strong&gt;Effort &amp;amp; Quick Wins&lt;/strong&gt;, &lt;strong&gt;Prerequisite Dependencies&lt;/strong&gt;, and &lt;strong&gt;Mental Energy Requirements&lt;/strong&gt;, spotlighting the single highest-yield item to tackle next with transparent explanations.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Accepts Natural-Language Task Dumps&lt;/strong&gt;: Lets users input thoughts naturally (&lt;em&gt;"Finish my DBMS assignment by tomorrow, prepare for Friday's exam, and spend 30 minutes replying to client emails"&lt;/em&gt;) and extracts actionable task structures with user review before saving.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Generates an Honest Daily Schedule ("Plan My Day")&lt;/strong&gt;: Slots tasks into realistic working hours, inserts recovery breaks, and honestly warns users when workloads exceed capacity rather than overpromising.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Protects Personal Privacy&lt;/strong&gt;: 100% of data remains stored locally on device with zero cloud telemetry.&lt;/li&gt;
&lt;/ol&gt;

&lt;h3&gt;
  
  
  Who Would Pay for It?
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Freelancers &amp;amp; Consultants&lt;/strong&gt;: Juggling multiple client deliverables with strict billable hour constraints who need to protect client confidentiality.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Students &amp;amp; Researchers&lt;/strong&gt;: Balancing exams, assignment deadlines, and study blocks who need automated workload pacing.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Small Business Owners&lt;/strong&gt;: Prioritizing operations and revenue-generating activities without expensive monthly SaaS subscription bloat.&lt;/li&gt;
&lt;/ul&gt;




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

&lt;p&gt;The complete source code is open-source under the MIT License on GitHub:&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/song-code-won" rel="noopener noreferrer"&gt;
        song-code-won
      &lt;/a&gt; / &lt;a href="https://github.com/song-code-won/PrioritiAI" rel="noopener noreferrer"&gt;
        PrioritiAI
      &lt;/a&gt;
    &lt;/h2&gt;
    &lt;h3&gt;
      PrioritiAI - Smart AI Task Planner with deterministic multi-factor priority scoring and local open-weight AI (Ollama/Qwen).
    &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;PrioritiAI — Smart AI Task Planner&lt;/h1&gt;
&lt;/div&gt;
&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;A privacy-first, open-weight AI productivity assistant that cuts through decision paralysis, calculates transparent priority rankings, and schedules realistic daily workflows.&lt;/strong&gt;&lt;/p&gt;
&lt;/blockquote&gt;
&lt;p&gt;&lt;a href="https://opensource.org/licenses/MIT" rel="nofollow noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/08cef40a9105b6526ca22088bc514fbfdbc9aac1ddbf8d4e6c750e3a88a44dca/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f4c6963656e73652d4d49542d626c75652e737667" alt="License: MIT"&gt;&lt;/a&gt;
&lt;a href="https://ollama.com" rel="nofollow noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/afc65ee3aec8b9ddd8b0271f4831ab7346e16d4233551288c9a06d4aa6c1c8ff/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f41492d4f6c6c616d612532302532462532305177656e253230322e352d707572706c652e737667" alt="Open Source AI"&gt;&lt;/a&gt;
&lt;a href="https://www.typescriptlang.org/" rel="nofollow noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/4ecbcc91a94b9190611390066f20064b709cfacf208b71b634e7d5a7c4dcbc5d/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f547970655363726970742d352e392d626c75652e737667" alt="TypeScript"&gt;&lt;/a&gt;
&lt;a href="https://react.dev/" rel="nofollow noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/ba3c159a0fd14b09005b75937b7f95f72b310fad55b0af7027dddb9cf95bfbc0/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f52656163742d31392d6379616e2e737667" alt="React 19"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;📖 1. Problem &amp;amp; Intended Users&lt;/h2&gt;
&lt;/div&gt;
&lt;p&gt;Students, freelancers, software developers, and small business operators frequently encounter task overload. When faced with 10–20 concurrent responsibilities across different contexts (deadlines, client emails, exam preparation, code reviews), humans experience &lt;strong&gt;decision fatigue&lt;/strong&gt; and default to either procrastinating or tackling low-impact tasks first.&lt;/p&gt;
&lt;p&gt;Most existing todo apps are mere static lists: they record what needs to be done, but provide zero guidance on &lt;strong&gt;what to do next and why&lt;/strong&gt;.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;PrioritiAI&lt;/strong&gt; bridges this gap:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Multi-Factor Priority Scoring&lt;/strong&gt;: Evaluates urgency, importance, effort, prerequisite dependencies, energy requirements, and impact with a 100% deterministic, explainable algorithm.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;"Your Next Best Task"&lt;/strong&gt;: Eliminates ambiguity by spotlighting the single highest-yield item to work on right now.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Open-Weight Local AI&lt;/strong&gt;: Ingests messy, natural-language…&lt;/li&gt;
&lt;/ol&gt;&lt;/div&gt;
  &lt;/div&gt;
  &lt;div class="gh-btn-container"&gt;&lt;a class="gh-btn" href="https://github.com/song-code-won/PrioritiAI" rel="noopener noreferrer"&gt;View on GitHub&lt;/a&gt;&lt;/div&gt;
&lt;/div&gt;


&lt;p&gt;&lt;strong&gt;Repository link:&lt;/strong&gt; &lt;a href="https://github.com/song-code-won/PrioritiAI" rel="noopener noreferrer"&gt;https://github.com/song-code-won/PrioritiAI&lt;/a&gt;&lt;/p&gt;




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

&lt;p&gt;PrioritiAI was built around a hybrid architecture pairing local open-weight AI with deterministic execution:&lt;/p&gt;

&lt;h3&gt;
  
  
  1. Open-Source AI Engine (Ollama + Qwen 2.5 / Mistral)
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Local Inference Runtime&lt;/strong&gt;: Integrates directly with &lt;a href="https://ollama.com" rel="noopener noreferrer"&gt;Ollama&lt;/a&gt; running locally at &lt;code&gt;http://localhost:11434&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Open-Weight Models&lt;/strong&gt;: Designed for &lt;code&gt;Qwen 2.5 (3B/7B)&lt;/code&gt; (Apache 2.0 license) and &lt;code&gt;Mistral 7B&lt;/code&gt; (Apache 2.0 license).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Task Parsing &amp;amp; Subtask Decomposition&lt;/strong&gt;: The local model interprets conversational task dumps, identifies entities (durations, deadlines, energy levels), and decomposes complex tasks into 3–5 actionable subtasks with time estimates.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Resilient Offline Fallback&lt;/strong&gt;: If Ollama is unavailable or offline, PrioritiAI automatically engages an embedded heuristic rule engine. Zero features or priority calculations ever break.&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  2. Deterministic Prioritization Scoring Formula
&lt;/h3&gt;

&lt;p&gt;Instead of relying on unpredictable black-box rankings, PrioritiAI uses a mathematically transparent priority formula ($0$ to $100$ pts):&lt;/p&gt;

&lt;p&gt;$$\text{Priority Score} = \min\left(100, \max\left(0, \text{Base Score} - \text{Dependency Penalty}\right)\right)$$&lt;/p&gt;

&lt;p&gt;Where:&lt;br&gt;
$$\text{Base Score} = (\text{Urgency} \times 0.40) + (\text{Importance} \times 0.35) + (\text{Effort} \times 0.15) + (\text{Impact} \times 0.10) + \text{Energy Bonus}$$&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Dependency Penalty (-80 pts)&lt;/strong&gt;: Tasks blocked by incomplete prerequisite tasks are penalized and moved to &lt;code&gt;Waiting on Something&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Quick Win Bonus (+40 pts)&lt;/strong&gt;: Sub-15 minute tasks receive an incentive boost to create early momentum.&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  3. Modern Tech Stack
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Frontend&lt;/strong&gt;: React 19, TypeScript 5.9, Tailwind CSS.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;State &amp;amp; Storage&lt;/strong&gt;: React Context + client-side LocalStorage persistence with full JSON export/import.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Testing&lt;/strong&gt;: Automated unit tests with Vitest covering priority scoring, blocker penalties, and the daily planner.&lt;/li&gt;
&lt;/ul&gt;




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

&lt;p&gt;Open innovation is what made PrioritiAI possible, practical, and ethically aligned with user trust:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;&lt;p&gt;&lt;strong&gt;Complete Data Sovereignty &amp;amp; Confidentiality&lt;/strong&gt;:&lt;br&gt;
With proprietary cloud APIs (OpenAI, Claude, etc.), every task—whether it's confidential client milestone notes or personal deadlines—must travel across the public internet to third-party data centers. Open-weight models running on consumer hardware ensure that zero bits of private task data ever leave the user's laptop.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;&lt;strong&gt;Zero Recurring API Bills for Side Hustlers&lt;/strong&gt;:&lt;br&gt;
For students and independent creators, paying $\$20$/month or per-token API charges for daily task management destroys the economics of a personal productivity tool. Open-weight models like Qwen 2.5 run on standard modern hardware at $\$0.00$ operational inference cost forever.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;&lt;strong&gt;Guaranteed Offline Resilience&lt;/strong&gt;:&lt;br&gt;
Productivity software cannot be brittle. Freelancers working on trains, flights, or with intermittent connections cannot afford to lose their task prioritizer. Open-source local AI delivers uninterrupted productivity anywhere on earth.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;&lt;strong&gt;Transparent, Explainable Logic&lt;/strong&gt;:&lt;br&gt;
Proprietary AI wrappers create black boxes where users don't know why a task was ranked where it was. By open-sourcing the deterministic scoring algorithm alongside local models, users have complete visibility and trust into why their next task was recommended.&lt;/p&gt;&lt;/li&gt;
&lt;/ol&gt;




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

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Primary Challenge&lt;/strong&gt;: Hacktoberfest Open-Source AI Challenge Week 2 — Side Hustle&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Open-Source AI &amp;amp; Local Inference Track&lt;/strong&gt;&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
    </item>
    <item>
      <title>CalorieWalk 🌿: Understand Your Meal, Then Step Away and Touch Grass</title>
      <dc:creator>Mayur P</dc:creator>
      <pubDate>Tue, 06 Oct 2026 04:34:53 +0000</pubDate>
      <link>https://dev.to/mayurponly/caloriewalk-understand-your-meal-then-step-away-and-touch-grass-5hj2</link>
      <guid>https://dev.to/mayurponly/caloriewalk-understand-your-meal-then-step-away-and-touch-grass-5hj2</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 commercial nutrition and food tracking applications are designed to hook you into endless screen time. They force you to weigh ingredients down to the single gram, scroll through overwhelming databases of brand names, and calculate obsessive deficit math (&lt;em&gt;"You ate 500 kcal, now spend 45 minutes on the treadmill to burn it off"&lt;/em&gt;). &lt;/p&gt;

&lt;p&gt;&lt;strong&gt;CalorieWalk&lt;/strong&gt; is built on the opposite philosophy: &lt;strong&gt;Screen time should be the shortest part of your day.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;CalorieWalk is a lightweight, privacy-first web application that allows you to photograph or quickly describe a meal. Using a locally running &lt;strong&gt;open-weight multimodal AI model&lt;/strong&gt;, it identifies individual meal components, estimates reasonable portions, and calculates approximate calories.&lt;/p&gt;

&lt;p&gt;The moment the meal is logged, the app invites you into &lt;strong&gt;"Touch Grass Mode"&lt;/strong&gt;—a dedicated, minimalist outdoor wellness flow with 10-, 20-, and 30-minute nature walk prompts. It features a distraction-free countdown timer designed so you can put your phone in your pocket, walk outside, look at the sky, and reconnect with the physical world.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Key principles:&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Zero Exercise Punishment:&lt;/strong&gt; CalorieWalk never prescribes workouts to "burn off" food. Outdoor strolls are suggested strictly as an independent mental and physical wellness break.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Honest Uncertainty:&lt;/strong&gt; Visual estimates are clearly labeled as approximate, acknowledging that photographic analysis cannot measure hidden cooking oils or exact sodium levels.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Privacy First:&lt;/strong&gt; Photos are processed locally with EXIF metadata stripped, never uploaded to third-party cloud servers, and discarded immediately after inference by default.&lt;/li&gt;
&lt;/ul&gt;




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

&lt;p&gt;The complete source code, Docker setup, and test suite are open source on GitHub:&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/song-code-won" rel="noopener noreferrer"&gt;
        song-code-won
      &lt;/a&gt; / &lt;a href="https://github.com/song-code-won/CalorieWalk" rel="noopener noreferrer"&gt;
        CalorieWalk
      &lt;/a&gt;
    &lt;/h2&gt;
    &lt;h3&gt;
      Understand your meal. Then go outside. Hacktoberfest Open-Source AI Challenge.
    &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;CalorieWalk 🌿 — Local AI Food &amp;amp; Calorie Estimator&lt;/h1&gt;
&lt;/div&gt;
&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Tagline:&lt;/strong&gt; Understand your meal. Then go outside.&lt;br&gt;
&lt;strong&gt;Challenge:&lt;/strong&gt; Hacktoberfest Open-Source AI Challenge: Week 1 — &lt;em&gt;“Touch Grass”&lt;/em&gt;&lt;/p&gt;
&lt;/blockquote&gt;
&lt;p&gt;CalorieWalk is a lightweight, privacy-first web application designed to help you quickly understand what you eat using a &lt;strong&gt;locally running open-weight AI model&lt;/strong&gt;, and then immediately step away from your device into the fresh air.&lt;/p&gt;

&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;🎯 The Problem &amp;amp; Philosophy&lt;/h2&gt;
&lt;/div&gt;
&lt;p&gt;Most commercial calorie tracking applications trap you in endless screen cycles: typing endless recipe variants, weighing grains to the single gram, and gamifying obsessive "calorie burn" calculations (&lt;em&gt;"You ate 500 kcal, now run on the treadmill for 45 minutes"&lt;/em&gt;).&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;CalorieWalk flips this paradigm:&lt;/strong&gt;&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Screen time is the shortest part of the experience:&lt;/strong&gt; Snap a meal photo or type a quick food note.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Local Open-Weight AI:&lt;/strong&gt; Processed right on your device—no cloud uploads, no subscription APIs.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Honest estimates with clear uncertainty:&lt;/strong&gt; Food…&lt;/li&gt;
&lt;/ol&gt;&lt;/div&gt;
  &lt;/div&gt;
  &lt;div class="gh-btn-container"&gt;&lt;a class="gh-btn" href="https://github.com/song-code-won/CalorieWalk" rel="noopener noreferrer"&gt;View on GitHub&lt;/a&gt;&lt;/div&gt;
&lt;/div&gt;


&lt;p&gt;&lt;strong&gt;Repository Link:&lt;/strong&gt; &lt;a href="https://github.com/song-code-won/CalorieWalk" rel="noopener noreferrer"&gt;https://github.com/song-code-won/CalorieWalk&lt;/a&gt;&lt;/p&gt;




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

&lt;p&gt;CalorieWalk was designed around an offline-capable, open-weight local inference pipeline without heavyweight cloud dependencies:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;p&gt;&lt;strong&gt;Open-Weight Multimodal AI Core:&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Powered by local open-weight Vision-Language Models running via &lt;strong&gt;Ollama&lt;/strong&gt; (e.g., &lt;code&gt;llava:latest&lt;/code&gt;, &lt;code&gt;llama3.2-vision:11b&lt;/code&gt;, or &lt;code&gt;qwen2-vl:7b&lt;/code&gt;).&lt;/li&gt;
&lt;li&gt;The vision service cleans and normalizes user uploads, strips GPS and camera EXIF metadata with &lt;strong&gt;Pillow&lt;/strong&gt;, and requests structured JSON outputs with bounded confidence scores.&lt;/li&gt;
&lt;li&gt;Built with strict schema validation and retry guards to reject hallucinated or invalid model structures.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;p&gt;&lt;strong&gt;Backend &amp;amp; Architecture:&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;FastAPI (Python 3.11):&lt;/strong&gt; High-performance asynchronous REST endpoints for image analysis, manual input, settings, and activity logs.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;SQLite:&lt;/strong&gt; Lightweight local persistence storing meal records and completed walks without needing external database servers.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Graceful Fallback Dataset:&lt;/strong&gt; A built-in reference dataset ensures users can still estimate standard meals even if their local model daemon is temporarily offline.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;p&gt;&lt;strong&gt;Frontend:&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Built with clean, accessible HTML5, responsive CSS, and vanilla JavaScript—zero bloated client-side JavaScript frameworks or compilation steps.&lt;/li&gt;
&lt;li&gt;Features an immersive &lt;strong&gt;Touch Grass screen&lt;/strong&gt; that replaces nutrition metrics with an outdoor countdown prompt.&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;Meal photographs and daily eating logs are deeply personal data. They reveal your daily routine, physical location, home interior, dining companions, and health habits. &lt;/p&gt;

&lt;p&gt;Relying on closed proprietary AI APIs (like OpenAI or cloud computer vision services) means funneling private personal photos to corporate servers for training and retention. &lt;/p&gt;

&lt;p&gt;Open-source and open-weight innovation fundamentally solves this:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Total Data Privacy:&lt;/strong&gt; Photos are analyzed directly on your own hardware. Not a single byte leaves the machine.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Offline Independence:&lt;/strong&gt; Once the open-weight model is pulled, your meal logging and outdoor companion function completely without an internet connection.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Auditable &amp;amp; Non-Extractive:&lt;/strong&gt; Anyone can verify that no hidden trackers or biometric harvesting occurs.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Zero Cost Barriers:&lt;/strong&gt; Users and developers avoid recurring token fees, subscription paywalls, or sudden proprietary API deprecations.&lt;/li&gt;
&lt;/ul&gt;




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

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Touch Grass Track:&lt;/strong&gt; Open-weight AI used specifically to reduce screen obsession and encourage users to step outside and experience nature.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Open-Source AI / Local Inference:&lt;/strong&gt; Built entirely around open-weight models running on local hardware without proprietary closed APIs.&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
    </item>
    <item>
      <title>NoteMate AI: A Local-First, Privacy-Preserving AI Note Assistant</title>
      <dc:creator>Mayur P</dc:creator>
      <pubDate>Fri, 02 Oct 2026 15:34:04 +0000</pubDate>
      <link>https://dev.to/mayurponly/notemate-ai-a-local-first-privacy-preserving-ai-note-assistant-1e44</link>
      <guid>https://dev.to/mayurponly/notemate-ai-a-local-first-privacy-preserving-ai-note-assistant-1e44</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;I built &lt;strong&gt;NoteMate AI&lt;/strong&gt; (&lt;em&gt;"Your thoughts, organized by AI. Your notes, under your control."&lt;/em&gt;) for my friend, who is currently managing a packed schedule of university lectures, academic study sessions, and team project meetings.&lt;/p&gt;

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

&lt;p&gt;My friend constantly ran into three major friction points:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Scattered, Messy Inputs:&lt;/strong&gt; Information was coming in from everywhere—spoken lectures, raw audio voice recordings, slide deck PDFs, and quickly typed scratch notes.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Cognitive Overload &amp;amp; Manual Revision:&lt;/strong&gt; Manually combing through hour-long lecture transcripts or 30-page slide PDFs to extract key definitions, action items, and flashcard revision questions was consuming hours of valuable study time.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Privacy Concerns &amp;amp; API Paywalls:&lt;/strong&gt; Commercial AI note apps require recurring monthly subscriptions ($15–$25/month) and send private voice recordings and proprietary course materials to third-party cloud servers.&lt;/li&gt;
&lt;/ol&gt;

&lt;h3&gt;
  
  
  What NoteMate AI Does
&lt;/h3&gt;

&lt;p&gt;NoteMate AI is a local, privacy-first personal AI note assistant designed specifically around my friend's workflow:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Instant AI Summarization:&lt;/strong&gt; In one click, local AI extracts a concise executive summary, bulleted key takeaways, important terminology, and actionable next steps—without altering or overwriting the original notes.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Voice-to-Notes:&lt;/strong&gt; Upload spoken audio recordings (&lt;code&gt;.mp3&lt;/code&gt;, &lt;code&gt;.wav&lt;/code&gt;, &lt;code&gt;.m4a&lt;/code&gt;). NoteMate AI transcribes them locally and automatically structures stream-of-consciousness speech into clean notes with clear headings and summaries.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;PDF-to-Notes:&lt;/strong&gt; Upload slide decks and reading materials. NoteMate AI extracts digital text and generates structured study guides while detecting scanned/image-only documents.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Strict Note-Scoped Q&amp;amp;A:&lt;/strong&gt; A dedicated Q&amp;amp;A assistant scoped strictly to the active note. If a fact isn't present in the note, the AI explicitly states that rather than hallucinating an answer.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Interactive AI Revision Quiz:&lt;/strong&gt; Generates a 5-question multiple-choice revision test with an interactive in-browser quiz runner, instant scoring, and explanation cards for active recall practice.&lt;/li&gt;
&lt;/ul&gt;




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

&lt;p&gt;The entire codebase is open-source and available on GitHub:&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/song-code-won" rel="noopener noreferrer"&gt;
        song-code-won
      &lt;/a&gt; / &lt;a href="https://github.com/song-code-won/notemate-ai" rel="noopener noreferrer"&gt;
        notemate-ai
      &lt;/a&gt;
    &lt;/h2&gt;
    &lt;h3&gt;
      NoteMate AI: A Personal AI Note-Taking Assistant powered by local open-weight AI (Gemma 3 &amp;amp; faster-whisper). Built for Hacktoberfest 2026.
    &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;NoteMate AI — A Personal AI Note-Taking Assistant&lt;/h1&gt;
&lt;/div&gt;
&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Tagline:&lt;/strong&gt; &lt;em&gt;Your thoughts, organized by AI. Your notes, under your control.&lt;/em&gt;&lt;br&gt;
Built for the &lt;strong&gt;Hacktoberfest 2026 DEV Challenge — Build for a Friend&lt;/strong&gt;.&lt;/p&gt;
&lt;/blockquote&gt;
&lt;p&gt;&lt;a href="https://www.python.org/downloads/" rel="nofollow noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/36cf3d0f7992a33a063d3833577d62204f8934d82b69874c086390608db4947c/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f707974686f6e2d332e31312b2d626c75652e737667" 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/15a792528b351b1abf3ded90cdf144879e6caeaf56600f03a612432e4517b7fd/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f466173744150492d302e3131302b2d677265656e2e737667" alt="FastAPI"&gt;&lt;/a&gt;
&lt;a href="https://ollama.com/" rel="nofollow noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/01d8744585039ac4392dac65ba57481da5aace01f2a1111c55b9c1d95ed56106/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f4f6c6c616d612d47656d6d615f332d6f72616e67652e737667" alt="Ollama"&gt;&lt;/a&gt;
&lt;a href="https://github.com/SYSTRAN/faster-whisper" rel="noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/6f07f7e56e36d9a3b4e3c6def9e8d0ecace03647fb0fd7cb771e7520c0c2b73f/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f5370656563682d2d746f2d2d546578742d6661737465722d2d776869737065722d707572706c652e737667" alt="Speech-to-Text"&gt;&lt;/a&gt;
&lt;a href="https://github.com/song-code-won/notemate-ai/LICENSE" rel="noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/fdf2982b9f5d7489dcf44570e714e3a15fce6253e0cc6b5aa61a075aac2ff71b/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f4c6963656e73652d4d49542d79656c6c6f772e737667" alt="License: MIT"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;1. Project Overview&lt;/h2&gt;
&lt;/div&gt;
&lt;p&gt;&lt;strong&gt;NoteMate AI&lt;/strong&gt; is a lightweight, simple, and functional AI-powered note-taking application designed to capture, organize, summarize, and retrieve knowledge effortlessly.&lt;/p&gt;
&lt;p&gt;Rather than being another generic AI chat wrapper, NoteMate AI is engineered as an active cognitive assistant that converts raw thoughts, recorded audio, and PDF documents into clean, structured notes—running &lt;strong&gt;100% locally&lt;/strong&gt; using open-weight AI models.&lt;/p&gt;

&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;2. The Problem &amp;amp; Who It Is Built For&lt;/h2&gt;
&lt;/div&gt;
&lt;div class="markdown-heading"&gt;
&lt;h3 class="heading-element"&gt;The Friend &amp;amp; The Use Case&lt;/h3&gt;

&lt;/div&gt;
&lt;p&gt;This project was built for my friend, who is actively balancing intense study sessions, university lectures, and collaborative team meetings.&lt;/p&gt;
&lt;div class="markdown-heading"&gt;
&lt;h3 class="heading-element"&gt;The Pain Points&lt;/h3&gt;

&lt;/div&gt;
&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Scattered Information:&lt;/strong&gt; Information arrives in unpredictable formats—spoken audio from lectures, lecture slide PDFs, and messy quick thoughts.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Cognitive Overload:&lt;/strong&gt; Manually reading…&lt;/li&gt;
&lt;/ol&gt;&lt;/div&gt;
  &lt;/div&gt;
  &lt;div class="gh-btn-container"&gt;&lt;a class="gh-btn" href="https://github.com/song-code-won/notemate-ai" rel="noopener noreferrer"&gt;View on GitHub&lt;/a&gt;&lt;/div&gt;
&lt;/div&gt;


&lt;p&gt;&lt;strong&gt;Repository:&lt;/strong&gt; &lt;a href="https://github.com/song-code-won/notemate-ai" rel="noopener noreferrer"&gt;https://github.com/song-code-won/notemate-ai&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  Tech Stack Overview
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Backend:&lt;/strong&gt; Python 3.11+, FastAPI, Uvicorn&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Local Open-Weight LLM:&lt;/strong&gt; Google Gemma 3 (&lt;code&gt;gemma3:1b&lt;/code&gt; / &lt;code&gt;gemma3:4b&lt;/code&gt;) running locally via &lt;strong&gt;Ollama&lt;/strong&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Speech-to-Text:&lt;/strong&gt; &lt;code&gt;faster-whisper&lt;/code&gt; (CTranslate2) running locally on CPU&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Document Processing:&lt;/strong&gt; PyMuPDF (&lt;code&gt;pymupdf&lt;/code&gt;)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Database:&lt;/strong&gt; SQLite with Write-Ahead Logging (WAL) mode for persistent storage&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Frontend:&lt;/strong&gt; Clean, responsive HTML5 / CSS3 / Vanilla JavaScript dashboard with split-view editor&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Testing &amp;amp; Containerization:&lt;/strong&gt; 19 automated Pytest cases, &lt;code&gt;Dockerfile&lt;/code&gt;, and &lt;code&gt;docker-compose.yml&lt;/code&gt;
&lt;/li&gt;
&lt;/ul&gt;




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

&lt;h3&gt;
  
  
  1. Open-Weight AI Architecture with Ollama &amp;amp; Gemma 3
&lt;/h3&gt;

&lt;p&gt;Rather than calling commercial cloud APIs like OpenAI or Anthropic, NoteMate AI's core intelligence is powered by &lt;strong&gt;Gemma 3&lt;/strong&gt; running locally through &lt;strong&gt;Ollama&lt;/strong&gt; (&lt;code&gt;http://localhost:11434&lt;/code&gt;).&lt;/p&gt;

&lt;p&gt;We designed a dedicated &lt;code&gt;AIService&lt;/code&gt; module that handles:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Structured prompt formatting for key-point and action-item extraction.&lt;/li&gt;
&lt;li&gt;Strict anti-hallucination prompting for the Q&amp;amp;A engine so answers remain grounded only in the user's note.&lt;/li&gt;
&lt;li&gt;JSON-formatted output parsing with robust regex fallbacks for 5-question MCQ generation.&lt;/li&gt;
&lt;li&gt;Dynamic fallback error reporting with actionable troubleshooting instructions if Ollama is paused or the model needs to be pulled.&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  2. Local Speech-to-Text with &lt;code&gt;faster-whisper&lt;/code&gt;
&lt;/h3&gt;

&lt;p&gt;For voice recordings, we integrated &lt;code&gt;faster-whisper&lt;/code&gt;, an optimized CTranslate2 reimplementation of OpenAI's Whisper model. By utilizing the quantized &lt;code&gt;tiny&lt;/code&gt; or &lt;code&gt;base&lt;/code&gt; weights on CPU, NoteMate AI transcribes audio files up to 4x faster than standard PyTorch models with minimal memory overhead, followed by an automatic Gemma 3 structuring pass.&lt;/p&gt;

&lt;h3&gt;
  
  
  3. PDF Parsing with PyMuPDF
&lt;/h3&gt;

&lt;p&gt;Text from multi-page PDFs is extracted directly using PyMuPDF. We implemented automated character-density detection so that if my friend uploads a scanned or image-only document, NoteMate AI flags it immediately and prompts for a digital PDF rather than failing silently.&lt;/p&gt;

&lt;h3&gt;
  
  
  4. Zero Data Loss with SQLite WAL
&lt;/h3&gt;

&lt;p&gt;To ensure notes survive system crashes and restarts without requiring a bulky database server, notes and AI summaries are stored persistently in a local SQLite database configured with &lt;code&gt;PRAGMA journal_mode=WAL;&lt;/code&gt; and indexed by modification timestamps.&lt;/p&gt;




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

&lt;p&gt;Building NoteMate AI with open-weight models and open-source tooling demonstrated why open innovation is critical for real-world personal software:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;&lt;p&gt;&lt;strong&gt;Complete Data Sovereignty &amp;amp; Privacy:&lt;/strong&gt;&lt;br&gt;&lt;br&gt;
Personal thoughts, course notes, and voice recordings never leave the laptop. For students and professionals dealing with confidential lectures, proprietary team discussions, or sensitive research, closed cloud APIs represent an unacceptable privacy trade-off. With Gemma 3 and faster-whisper, inference happens 100% on localhost.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;&lt;strong&gt;Democratizing Accessibility (Zero Subscription Barriers):&lt;/strong&gt;&lt;br&gt;&lt;br&gt;
My friend shouldn't have to budget $20/month just to summarize study notes or prepare for an exam. Open models turn consumer laptops into self-sufficient AI workstations without credit card requirements or surprise monthly billing.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;&lt;strong&gt;True Model Flexibility:&lt;/strong&gt;&lt;br&gt;&lt;br&gt;
Because the system communicates with Ollama through open standards, my friend isn't locked into a single provider. With a single environment variable change, the model can switch between &lt;code&gt;gemma3:1b&lt;/code&gt; for battery conservation on an airplane, &lt;code&gt;gemma3:4b&lt;/code&gt; for deep revision, or open models like &lt;code&gt;llama3&lt;/code&gt; or &lt;code&gt;mistral&lt;/code&gt;.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;&lt;strong&gt;Auditability and Transparency:&lt;/strong&gt;&lt;br&gt;&lt;br&gt;
Every prompt, processing pipeline, and storage schema is open and verifiable. There are no black-box system prompts or hidden telemetry collecting personal data.&lt;/p&gt;&lt;/li&gt;
&lt;/ol&gt;




&lt;h2&gt;
  
  
  Real User Feedback (From My Friend)
&lt;/h2&gt;

&lt;p&gt;After testing NoteMate AI with actual course materials and lecture recordings on an Intel i7 laptop (16GB RAM):&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;em&gt;"The voice-to-notes workflow saved me nearly half an hour after a two-hour lecture. Instead of re-listening to messy audio, having the transcript cleaned up with headings and action items in one place made review instantaneous.&lt;/em&gt;  &lt;/p&gt;

&lt;p&gt;&lt;em&gt;The revision quiz was surprisingly fun and accurate—it asked questions about the exact concepts covered in the notes rather than generic trivia."&lt;/em&gt;&lt;/p&gt;
&lt;/blockquote&gt;




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

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Primary Track:&lt;/strong&gt; Build for a Friend&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Open-Source AI / Open-Weight Model Track:&lt;/strong&gt; Powered by Google Gemma 3 via Ollama &amp;amp; faster-whisper&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>devchallenge</category>
      <category>weekendchallenge</category>
      <category>hf26challenge</category>
    </item>
    <item>
      <title>Hello there</title>
      <dc:creator>Mayur P</dc:creator>
      <pubDate>Thu, 01 Oct 2026 15:56:51 +0000</pubDate>
      <link>https://dev.to/mayurponly/hello-there-37ec</link>
      <guid>https://dev.to/mayurponly/hello-there-37ec</guid>
      <description>&lt;p&gt;Hello everyone!  Excited to join this amazing developer community, learn new things, share ideas, and connect with fellow developers! &lt;/p&gt;

</description>
    </item>
  </channel>
</rss>
