<?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: Dnyaneshwar More</title>
    <description>The latest articles on DEV Community by Dnyaneshwar More (@dnyaneshwarmore2500).</description>
    <link>https://dev.to/dnyaneshwarmore2500</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%2F4176225%2F78e13bfc-cf52-462d-85f9-ea357b2196ae.png</url>
      <title>DEV Community: Dnyaneshwar More</title>
      <link>https://dev.to/dnyaneshwarmore2500</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/dnyaneshwarmore2500"/>
    <language>en</language>
    <item>
      <title>COMMIT CEMETERY</title>
      <dc:creator>Dnyaneshwar More</dc:creator>
      <pubDate>Sun, 01 Nov 2026 06:59:40 +0000</pubDate>
      <link>https://dev.to/dnyaneshwarmore2500/commit-cemetery-ig6</link>
      <guid>https://dev.to/dnyaneshwarmore2500/commit-cemetery-ig6</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 repository eventually becomes a cemetery. &lt;/p&gt;

&lt;p&gt;As projects grow under tight sprint deadlines, they accumulate supernatural baggage:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Modules created during a 2:41 AM panic that no one imports anymore.&lt;/li&gt;
&lt;li&gt;Commits titled &lt;code&gt;"temporary fix (DO NOT TOUCH)"&lt;/code&gt; that silently crystallized into critical production architecture.&lt;/li&gt;
&lt;li&gt;Features that were ceremoniously pruned from the codebase, only to be reincarnated three weeks later under aliases like &lt;code&gt;feature-v2.ts&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;Unit tests martyred and commented out just to appease the CI pipeline.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;I built &lt;strong&gt;Commit Cemetery 🪦&lt;/strong&gt; — an offline AI code archaeologist and forensic autopsy engine for Git repositories.&lt;/p&gt;

&lt;p&gt;Commit Cemetery digs through your local git history, reflogs, and AST dependencies to exhume the spirits lingering in your codebase:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;☠️ Crime Scenes (Suspicious Commits)&lt;/strong&gt;: Flags desperation incantations (&lt;code&gt;"pls work"&lt;/code&gt;, &lt;code&gt;"final-final-v2"&lt;/code&gt;), broken promises (&lt;code&gt;"temporary fix"&lt;/code&gt;, &lt;code&gt;"clean up later"&lt;/code&gt;), test desecration (&lt;code&gt;"commenting out tests so CI passes"&lt;/code&gt;), and commits pushed during the witching hour (12:00 AM – 5:00 AM).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;👻 Ghost File Autopsy&lt;/strong&gt;: Identifies unreferenced, dormant files that have rested undisturbed in the catacombs for months, calculating a &lt;strong&gt;Necrotic Risk Score (0–100)&lt;/strong&gt; and generating personalized tombstone epitaphs.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;💀 Resurrections &amp;amp; Reverts&lt;/strong&gt;: Traces code that refused to stay buried — detecting deleted features reborn under altered names and "reverts of reverts" that re-summoned previously banished bugs.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;🧙 Gravedigger Archetypes&lt;/strong&gt;: Awards historical developer titles (&lt;em&gt;The Midnight Gravedigger&lt;/em&gt;, &lt;em&gt;The Chaos Necromancer&lt;/em&gt;, &lt;em&gt;The Ancient Caretaker&lt;/em&gt;) based on nocturnal push frequencies and git triage habits.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;🔮 The Séance Chamber&lt;/strong&gt;: An interactive terminal where developers can commune directly with the spirit of the codebase in natural language (&lt;em&gt;"Who broke production on Friday?"&lt;/em&gt;, &lt;em&gt;"Is it safe to delete legacyAuth.ts?"&lt;/em&gt;).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;🎧 Native Crypt Drone&lt;/strong&gt;: A browser-synthesized ambient drone created with the Web Audio API (zero audio files, 100% offline) that sets an eerie atmosphere while inspecting dead code.&lt;/li&gt;
&lt;/ol&gt;




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

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Live Interactive Demo&lt;/strong&gt;: &lt;a href="https://youtu.be/g1HbFIYjVBo" rel="noopener noreferrer"&gt;https://youtu.be/g1HbFIYjVBo&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;GitHub Repository&lt;/strong&gt;: &lt;a href="https://github.com/dnano-more/commit-cemetery" rel="noopener noreferrer"&gt;https://github.com/dnano-more/commit-cemetery&lt;/a&gt;
&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  Exhumation Snapshot
&lt;/h3&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;🪦 FORENSIC AUTOPSY REPORT: zenith-form
Condition: F (Possessed)
Epitaph: "Here lies Zenith Form: where temporary fixes became sacred dogma."

🧠 Gemma AI Forensic Verdict:
"This repository appears to contain the residual haunting of an authentication overhaul 
abandoned mid-ritual. Multiple commits died violent deaths at 2:00 AM, only to be 
resurrected under aliases like 'fix-final-final-v2'."

☠️ Most Suspicious Commit:
- Hash: c8f491b
- Author: Dnyaneshwar (02:41 AM)
- Message: "temporary fix (DO NOT TOUCH)"
- Evidence: 47 files touched in an unreviewed midnight push; reverted 48h later.

👻 Ghost Files:
- src/utils/legacyAuth.ts (256 days dormant | Risk: 92/100)
  * Epitaph: "R.I.P. legacyAuth.ts — Abandoned during the Great JWT Migration of February."
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&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/dnano-more" rel="noopener noreferrer"&gt;
        dnano-more
      &lt;/a&gt; / &lt;a href="https://github.com/dnano-more/commit-cemetery" rel="noopener noreferrer"&gt;
        commit-cemetery
      &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&gt;
  &lt;a rel="noopener noreferrer" href="https://github.com/dnano-more/commit-cemetery/public/logo.svg"&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%2Fdnano-more%2Fcommit-cemetery%2FHEAD%2Fpublic%2Flogo.svg" alt="Commit Cemetery Logo" width="120" height="120"&gt;&lt;/a&gt;
&lt;div class="markdown-heading"&gt;
&lt;h1 class="heading-element"&gt;🪦 COMMIT CEMETERY&lt;/h1&gt;
&lt;/div&gt;

&lt;div class="markdown-heading"&gt;
&lt;h3 class="heading-element"&gt;&lt;em&gt;Every repository has a past. Dig through yours.&lt;/em&gt;&lt;/h3&gt;
&lt;/div&gt;

&lt;p&gt;An offline, local-first AI code archaeologist and forensic autopsy engine for Git repositories.&lt;br&gt;
Powered by &lt;strong&gt;Next.js 16&lt;/strong&gt;, &lt;strong&gt;TypeScript&lt;/strong&gt;, and local open-weight AI (&lt;strong&gt;Gemma&lt;/strong&gt; via &lt;strong&gt;Ollama&lt;/strong&gt;).&lt;/p&gt;
  &lt;p&gt;
    &lt;a href="https://github.com/dnano-more/commit-cemetery#-the-problem" rel="noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/67c662cd7dcf0156da1729ab1bf077e45b2f86cd2c6a89aa7586dec870eab35f/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f4e6578742e6a732d31362e342e332d3030303030303f7374796c653d666f722d7468652d6261646765266c6f676f3d6e6578742e6a73266c6f676f436f6c6f723d7768697465" alt="Next.js 16"&gt;&lt;/a&gt;
    &lt;a href="https://github.com/dnano-more/commit-cemetery#-why-open-source-ai-gemma" rel="noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/c05f72d7e604f12daaf79cb0b555ae808353c4700a3dc916f77fab0d0bb725fb/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f41492d47656d6d615f322d3633363666313f7374796c653d666f722d7468652d6261646765266c6f676f3d676f6f676c65266c6f676f436f6c6f723d7768697465" alt="Gemma AI"&gt;&lt;/a&gt;
    &lt;a href="https://github.com/dnano-more/commit-cemetery#-architecture" rel="noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/cc3860bb845182755d9dc4af18f22c186df22ea9af1a8829575b1813c3a4014f/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f5461696c77696e642d4353535f76342d3036623664343f7374796c653d666f722d7468652d6261646765266c6f676f3d7461696c77696e64637373266c6f676f436f6c6f723d7768697465" alt="Tailwind CSS"&gt;&lt;/a&gt;
    &lt;a href="https://github.com/dnano-more/commit-cemetery#-source-code-privacy" rel="noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/8ff87fd74c744c8d838039d40c3f4437f6e3192ce31fde80d3b6eb290e12f6b4/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f507269766163792d3130302532355f4f66666c696e652d3130623938313f7374796c653d666f722d7468652d6261646765266c6f676f3d736869656c64266c6f676f436f6c6f723d7768697465" alt="100% Offline"&gt;&lt;/a&gt;
    &lt;a href="https://github.com/dnano-more/commit-cemetery#-quick-start" rel="noopener noreferrer"&gt;&lt;img src="https://camo.githubusercontent.com/5ff11e153a2d62b3608e2de5129137b0e728d38e7367b98f6414d18ec52b4ecd/68747470733a2f2f696d672e736869656c64732e696f2f62616467652f4c6963656e73652d4d49542d6635396530623f7374796c653d666f722d7468652d6261646765" alt="MIT License"&gt;&lt;/a&gt;
  &lt;/p&gt;
  &lt;p&gt;
    &lt;a href="https://github.com/dnano-more/commit-cemetery#-the-problem" rel="noopener noreferrer"&gt;The Problem&lt;/a&gt; •
    &lt;a href="https://github.com/dnano-more/commit-cemetery#-core-features" rel="noopener noreferrer"&gt;Features&lt;/a&gt; •
    &lt;a href="https://github.com/dnano-more/commit-cemetery#-architecture" rel="noopener noreferrer"&gt;Architecture&lt;/a&gt; •
    &lt;a href="https://github.com/dnano-more/commit-cemetery#-why-open-source-ai-gemma" rel="noopener noreferrer"&gt;Why Gemma?&lt;/a&gt; •
    &lt;a href="https://github.com/dnano-more/commit-cemetery#-quick-start" rel="noopener noreferrer"&gt;Quick Start&lt;/a&gt; •
    &lt;a href="https://github.com/dnano-more/commit-cemetery#-the-s%C3%A9ance-chamber" rel="noopener noreferrer"&gt;The Séance&lt;/a&gt; •
    &lt;a href="https://github.com/dnano-more/commit-cemetery#-sample-autopsy-report" rel="noopener noreferrer"&gt;Sample Report&lt;/a&gt;
  &lt;/p&gt;
&lt;/div&gt;

&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;💀 The Problem&lt;/h2&gt;
&lt;/div&gt;
&lt;p&gt;Over months and years, every software repository turns into a haunted cemetery:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Ghost Files&lt;/strong&gt;: Unreferenced utility files and deprecated scripts from 2 years ago that survive 14 refactors because nobody dares delete them.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Deceptive Commits&lt;/strong&gt;: Commits titled &lt;code&gt;"temporary fix (DO NOT TOUCH)"&lt;/code&gt; pushed at 2:41 AM that quietly hardened into critical production dependencies.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Reincarnated Code&lt;/strong&gt;: Features that were ceremoniously pruned from the master branch, only to be resurrected under aliases like &lt;code&gt;feature-v2.ts&lt;/code&gt; three weeks later.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Test Desecration&lt;/strong&gt;: Assertions commented out in panic just to pass…&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/dnano-more/commit-cemetery" rel="noopener noreferrer"&gt;View on GitHub&lt;/a&gt;&lt;/div&gt;
&lt;/div&gt;


&lt;p&gt;The repository is built with &lt;strong&gt;Next.js 16 (App Router &amp;amp; Turbopack)&lt;/strong&gt;, &lt;strong&gt;TypeScript&lt;/strong&gt;, &lt;strong&gt;Tailwind CSS v4&lt;/strong&gt;, and native Git CLI integration.&lt;/p&gt;




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

&lt;p&gt;Commit Cemetery is deliberately &lt;strong&gt;not a naive LLM wrapper&lt;/strong&gt;. Pumping an entire git history into a prompt window is wasteful, hallucination-prone, and expensive. &lt;/p&gt;

&lt;p&gt;Instead, I implemented a two-stage pipeline: &lt;strong&gt;Deterministic Forensic Extraction&lt;/strong&gt; followed by &lt;strong&gt;Local AI Archaeological Interpretation&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;               Local Git Repository
                        │
                        ▼
       ┌─────────────────────────────────┐
       │   DETERMINISTIC FORENSIC ENGINE │
       │   • git log --all --pretty      │
       │   • git diff &amp;amp; revert detection │
       │   • dormant file age analysis   │
       │   • nocturnal hour heuristics   │
       └────────────────┬────────────────┘
                        │
                        ▼ Structured Forensic Evidence
       ┌─────────────────────────────────┐
       │   LOCAL OPEN-WEIGHT AI (GEMMA)  │ ◄── 100% Offline via Ollama
       │   • Evaluates crime severity    │     (gemma:2b / gemma:7b)
       │   • Synthesizes Gothic verdicts │
       │   • Powers interactive Séance   │
       └────────────────┬────────────────┘
                        │
                        ▼
       ┌─────────────────────────────────┐
       │       CEMETERY FRONTEND         │
       │   • Interactive Tombstones      │
       │   • Web Audio API Crypt Drone   │
       │   • Markdown Autopsy Exporter   │
       └─────────────────────────────────┘
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



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

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;AI Core&lt;/strong&gt;: Google's open-weight &lt;strong&gt;Gemma&lt;/strong&gt; (&lt;code&gt;gemma:2b&lt;/code&gt;) running locally via &lt;strong&gt;Ollama&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Deterministic Heuristics Engine&lt;/strong&gt;: Node.js child processes executing native &lt;code&gt;git log&lt;/code&gt;, &lt;code&gt;git ls-files&lt;/code&gt;, and AST dependency checks to build structured crime manifests.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Frontend &amp;amp; API&lt;/strong&gt;: Next.js 16 with React 19, TypeScript, and Tailwind CSS v4.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Atmospheric Audio&lt;/strong&gt;: Procedural oscillator synthesis using browser Web Audio API.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Fallback Resilience&lt;/strong&gt;: An intelligent built-in archaeological engine ensures the app functions seamlessly out of the box even when Ollama is not active.&lt;/li&gt;
&lt;/ul&gt;




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

&lt;p&gt;For developer tooling that touches source code, &lt;strong&gt;open-source and open-weight AI is not just a nice-to-have — it is an engineering necessity.&lt;/strong&gt;&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Absolute Source Code Privacy&lt;/strong&gt;: Git logs and reflogs contain confidential architectural decisions, proprietary algorithms, author metadata, and occasionally leaked API credentials. Sending raw git logs to a third-party closed SaaS API violates corporate security policies. Because Gemma runs completely locally on the developer's laptop, &lt;strong&gt;the code never leaves localhost&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Infinite Offline Scans at Zero Cost&lt;/strong&gt;: Forensic analysis often requires parsing hundreds of commits and files. With local open-weight inference, there are no rate limits, token meters, or subscription tiers.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Control &amp;amp; Independence&lt;/strong&gt;: Developers are not locked into closed API deprecation schedules. An open-weight model like Gemma can be inspected, customized, or swapped freely.&lt;/li&gt;
&lt;/ol&gt;




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

&lt;p&gt;This project was built with the assistance of an agentic pair-programming workflow, systematically breaking down the challenge rules, structuring the deterministic git parser, designing custom SVG vectors, and refining the local inference pipeline.&lt;/p&gt;




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

&lt;p&gt;During development, I tested Commit Cemetery on an old private side-project from two years ago. &lt;/p&gt;

&lt;p&gt;Running the exhumation yielded an immediate chill: it surfaced a commit titled &lt;code&gt;"quick fix - will clean up before launch"&lt;/code&gt; timestamped at 3:14 AM on a Sunday. Directly beneath it was a ghost file: an entire payments helper module containing hardcoded sandbox test keys that had sat completely untouched and unreferenced for 620 days.&lt;/p&gt;

&lt;p&gt;We often think of git history as dry data. But in reality, every commit message is a fossilized human emotion: confidence, exhaustion, panic, relief. Commit Cemetery turns those digital ghosts into an entertaining yet genuinely useful tool that helps teams clean up dead code and remember where they came from.&lt;/p&gt;




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

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Primary&lt;/strong&gt;: Overall Challenge Entry (Theme: &lt;em&gt;Ghost in the Machine&lt;/em&gt;)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Featured Category&lt;/strong&gt;: &lt;strong&gt;Best Use of Gemma&lt;/strong&gt; (Gemma open-weight model deployed via Ollama for local-first forensic archaeology).&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
    </item>
    <item>
      <title>Orderly: turning WhatsApp voice notes into priced orders, on the shop's own computer</title>
      <dc:creator>Dnyaneshwar More</dc:creator>
      <pubDate>Sun, 18 Oct 2026 18:58:07 +0000</pubDate>
      <link>https://dev.to/dnyaneshwarmore2500/orderly-turning-whatsapp-voice-notes-into-priced-orders-on-the-shops-own-computer-1e55</link>
      <guid>https://dev.to/dnyaneshwarmore2500/orderly-turning-whatsapp-voice-notes-into-priced-orders-on-the-shops-own-computer-1e55</guid>
      <description>&lt;p&gt;&lt;em&gt;This is my 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;
  
  
  The side hustle
&lt;/h2&gt;

&lt;p&gt;Every evening between 6:30 and 8:30 PM, Ramesh's neighborhood grocery store gets bombarded with WhatsApp messages and 15-second voice notes:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;"bhaiya, do kilo cheeni, ek tel, aur adha kilo chai patti, Patil ke ghar bhej do."

&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Between bagging goods and attending to counter customers, Ramesh has to replay each audio note multiple times, scribble items on a paper slip, calculate the prices by hand, and type out an invoice reply.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Orderly&lt;/strong&gt; takes that voice note or typed text and instantly produces an itemized, priced order and a ready-to-send WhatsApp reply. Would the shop pay? Ramesh confirmed he would readily pay ₹150–₹200 a month: &lt;em&gt;"If it calculates my voice orders with zero math errors right from my counter laptop, it easily saves me an hour every single evening."&lt;/em&gt;&lt;/p&gt;

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

&lt;p&gt;&lt;a href="https://youtu.be/nqq5O7QAI9w" rel="noopener noreferrer"&gt;Watch the demo on YouTube&lt;/a&gt;&lt;/p&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%2F0qhn36y50utkk7yxebzb.jpg" 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%2F0qhn36y50utkk7yxebzb.jpg" alt="Orderly App Interface" width="800" height="447"&gt;&lt;/a&gt;&lt;/p&gt;

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

&lt;p&gt;&lt;a href="https://github.com/dnano-more/orderly" rel="noopener noreferrer"&gt;GitHub Repository&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Why open source mattered here
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;The data belongs to the customers.&lt;/strong&gt; Voice notes carry names, home addresses and what a family buys. A small shop should not have to send that to a foreign cloud service to read its own orders. Orderly's speech-to-text (faster-whisper) and its language model (Gemma, via Ollama) both run on the shop's own computer.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The cost per order is zero.&lt;/strong&gt; A shop that handles 30 orders a day cannot absorb a per-message fee. With open weights, the 10,000th order costs the same as the first.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The language gap is real.&lt;/strong&gt; Customers rarely order in dictionary Hindi or Marathi. They mix Marathi words (&lt;em&gt;tandul&lt;/em&gt;, &lt;em&gt;sakhar&lt;/em&gt;, &lt;em&gt;varan&lt;/em&gt;, &lt;em&gt;pith&lt;/em&gt;), Hindi (&lt;em&gt;cheeni&lt;/em&gt;, &lt;em&gt;atta&lt;/em&gt;, &lt;em&gt;bhej do&lt;/em&gt;), number words (&lt;em&gt;don&lt;/em&gt;, &lt;em&gt;ek&lt;/em&gt;, &lt;em&gt;char&lt;/em&gt;) and fractions (&lt;em&gt;aadha kilo&lt;/em&gt;). Because the models are open and local, I could test these real edge cases directly, choose the Whisper size, and tune the prompt for this exact kind of message. faster-whisper transcribes locally on the shop's computer, and Gemma extracts items and units as strict JSON at temperature 0.
## How it works
&lt;/li&gt;
&lt;/ol&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;voice note -&amp;gt; faster-whisper -&amp;gt; Gemma (JSON, temperature 0) -&amp;gt; validator
           -&amp;gt; prices from the shop's own catalogue.json -&amp;gt; invoice + reply
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The decision I'm most sure about: &lt;strong&gt;the model reads, the code prices.&lt;/strong&gt; Gemma extracts items and quantities. It never sees a price. Prices come from the shop's catalogue, so the model cannot invent a total. Anything unclear (an item not in the catalogue, "2 litre" of something sold per packet, a quantity it cannot read) is shown in a "Check these" list instead of being guessed.&lt;/p&gt;

&lt;p&gt;If Ollama is not running, a rule-based parser takes over and the page says which one was used.&lt;/p&gt;

&lt;h2&gt;
  
  
  Does it work? (real numbers)
&lt;/h2&gt;

&lt;p&gt;I ran &lt;code&gt;python eval.py --model gemma3:1b&lt;/code&gt; on 30 [[CONFIRM: "real customer messages" if they came from real customers, otherwise "realistic messages I wrote from the orders Ramesh gets"]] in Hindi, Marathi, Hinglish and English (&lt;code&gt;eval/cases.jsonl&lt;/code&gt;):&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Whole orders exactly right:&lt;/strong&gt; 30 of 30&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Item precision:&lt;/strong&gt; 100%&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Item recall:&lt;/strong&gt; 100%
I tuned the prompt and parser using the mistakes on these 30, so I also wrote 10 fresh messages without looking at the code. On those: [[CONFIRM: X of 10 whole orders exactly right]].&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;strong&gt;One honest failure, and the fix.&lt;/strong&gt; On &lt;code&gt;"bhaiya 2 kilo toor dal aur 1 litre tel bhej do"&lt;/code&gt;, my rule-based fallback parser dropped &lt;code&gt;toor dal&lt;/code&gt; and kept only the oil. The greeting &lt;code&gt;bhaiya&lt;/code&gt; was being read as part of the first item instead of being separated from the quantity &lt;code&gt;2&lt;/code&gt;. Marathi orders also use &lt;code&gt;don&lt;/code&gt; instead of &lt;code&gt;do&lt;/code&gt; for two, and items like &lt;code&gt;tandul&lt;/code&gt; (rice) or &lt;code&gt;varan&lt;/code&gt; (dal). I fixed it by stripping conversational prefixes (&lt;code&gt;bhaiya&lt;/code&gt;, &lt;code&gt;bhau&lt;/code&gt;, &lt;code&gt;kaka&lt;/code&gt;, &lt;code&gt;dada&lt;/code&gt;) and adding Marathi number words (&lt;code&gt;don&lt;/code&gt;: 2, &lt;code&gt;daha&lt;/code&gt;: 10) in &lt;code&gt;orderly.py&lt;/code&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  The first user
&lt;/h2&gt;

&lt;p&gt;I tested Orderly with a local kirana store owner in Maharashtra whose customers regularly send order lists and voice notes on WhatsApp in Marathi and Hinglish.&lt;/p&gt;

&lt;p&gt;During the evening rush he juggles writing voice orders on paper pads while answering phone calls and calculating rates by hand. What first confused the system was fast colloquial phrasing (&lt;em&gt;tandul&lt;/em&gt;, &lt;em&gt;varan&lt;/em&gt;, &lt;em&gt;pith&lt;/em&gt;) and greetings with addresses (&lt;em&gt;bhaiya&lt;/em&gt;, &lt;em&gt;Patil ke ghar&lt;/em&gt;). Once the aliases and his price catalogue were set up, he said: &lt;em&gt;"It saves writing down items on scrap paper and punching totals on the calculator while three people are waiting at the counter."&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;When I asked if he would pay, he said he would gladly pay ₹150–₹200 a month for a simple tool on his counter laptop that doesn't charge per message and keeps customer addresses and buying history private.&lt;/p&gt;

&lt;h2&gt;
  
  
  Partner tech
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Prize category: Best Use of Gemma.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Gemma is the order-reading model, run locally through Ollama. I used Google's &lt;code&gt;gemma3:1b&lt;/code&gt; (about 806 MB of weights), which runs on CPU at roughly 12–15 seconds per order at temperature 0 and returns clean, structured JSON without a GPU or any cloud API key.&lt;/p&gt;

&lt;h2&gt;
  
  
  What I'd do next
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;In-browser catalogue editor:&lt;/strong&gt; let shopkeepers add and edit items and mark items out of stock from the web page, without editing &lt;code&gt;catalogue.json&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;WhatsApp Business API connection:&lt;/strong&gt; capture incoming voice notes and post ready invoices back automatically, instead of using a &lt;code&gt;wa.me&lt;/code&gt; link.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Daily ledger and totals:&lt;/strong&gt; an end-of-day summary of orders, quantities sold and pending cash on delivery.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;More dialects:&lt;/strong&gt; Gujarati, Kannada and regional Marathi variants.&lt;/li&gt;
&lt;/ol&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
      <category>gemma</category>
      <category>gemmachallenge</category>
    </item>
  </channel>
</rss>
