DEV Community: Faheem Musthafa P

CodeNecromancer: Resurrecting Dead Repositories with Open-Weight Gemma

Faheem Musthafa P — Fri, 30 Oct 2026 02:28:20 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge: Week 4.

💀 Ghost in the Machine: What I Built

Every developer knows the sorrow of wandering through the open-source graveyard. Repositories that once powered thousands of applications sit abandoned in digital purgatory—strangled by bitrot, deprecated packages, vulnerable cryptographic primitives, and broken CI systems like Travis CI.

For Hacktoberfest 2026: Week 4 (Ghost in the Machine), I built CodeNecromancer: an autonomous digital mortician and repository resurrection agent powered by Google Gemma (open-weight) running locally on-device.

CodeNecromancer does three things:

The Autopsy Chamber (codenecromancer autopsy): Inspects a repository's git burial ground, calculates a Necro-Score (0–100), audits rotting dependencies, and channels Gemma to issue a Certificate of Digital Death with deep technical post-mortem analysis.
The Séance Circle (codenecromancer seance): Holds an interactive séance with the Ghost of the Original Maintainer, allowing you to interrogate the ghost about their architecture, unmerged plans, and regrets.
The Resurrection Ritual (codenecromancer resurrect): Autonomously mutates rotting manifests, synthesizes modern GitHub Actions v4 CI workflows, refactors deprecated syntax (such as Python 2 or obsolete APIs), and generates a commemorative RESURRECTION.md on a fresh git branch.

It includes both a Rich Gothic Terminal CLI and a web portal with dark ethereal styling.

🔮 Why Open Innovation Matters

Code bases are among the most sensitive intellectual property developers create. In this project, relying on open-weight models (Google Gemma via Ollama) wasn't just a design choice—it was the only ethical and practical architecture:

Zero Source Code Exfiltration: When auditing an internal or private abandoned repository, sending entire git histories, environment variables, and proprietary code to cloud LLMs violates security policies. With open-weight Gemma running locally on an RTX GPU, zero bytes leave the machine.
No Token Tax on Deep AST Analysis: Performing full-repo dependency graphs, changelog forensics, and multi-file code refactoring consumes hundreds of thousands of tokens. Running on-device is 100% free, unlimited, and runs even with the lights and internet turned off.
Perpetual Digital Preservation: Proprietary APIs alter model weights, introduce new safety guardrails that break syntax parsing, or shut down entirely. Open-weight models guarantee that software archaeology tools remain functional decades into the future.

⚡ Demo & Screenshots

Terminal Autopsy Output

╭──────────────────────────────────────────────────────────────────────────────╮
│                       📜 CERTIFICATE OF DIGITAL DEATH                        │
│ ┌───────────────────────────┬──────────────────────────────────────────────┐ │
│ │ Repository Subject        │ zombie_microservice                          │ │
│ │ Necro-Score               │ 100 / 100 (Fossilized Lich)                  │ │
│ │ Time in Purgatory         │ 3,487 days (Last breath: 2017-04-12)         │ │
│ │ Defunct CI/CD             │ Travis CI (Defunct/Abandoned)                │ │
│ │ Dead / Vulnerable         │ • flask==0.12.2 • pycrypto==2.6.1            │ │
│ │ Packages                  │ • requests==2.11.1 • nose==1.3.7             │ │
│ │ ⚰️ Official Cause of Death │ Succumbed to digital entropy, rotting in... │ │
│ └───────────────────────────┴──────────────────────────────────────────────┘ │
╰──────────────────────────────────────────────────────────────────────────────╯

The Séance with the Ghost of the Maintainer

User: "Why did you abandon this repository?"

Ghost of zombie_microservice: "I built this thing to serve, but time turned my function into a mausoleum. My dependencies are whispers of vulnerability, and my CI/CD is a forgotten crypt. The pull requests stopped coming in 2017..."

🛠️ Partner Technology & Tracks Entered

Best Use of Gemma ($200 Featured Track):
- Powered by Google Gemma (gemma4:e4b / gemma:2b) running via local Ollama.
- Gemma handles:
- The qualitative post-mortem autopsy analysis and cause-of-death diagnosis.
- Dynamic roleplay as the Departed Maintainer in the séance chat.
- Code modernization refactoring (transforming legacy syntax into modern standard Python).

🧪 Testing Instructions for Judges

To run CodeNecromancer locally:

# 1. Clone & install
git clone https://github.com/FM9447/CodeNecromancer.git
cd CodeNecromancer
pip install -e .

# 2. Ensure Ollama is running with Gemma
ollama pull gemma4:e4b  # or gemma:2b

# 3. Autopsy the included 2017 zombie microservice
codenecromancer autopsy examples/zombie_microservice

# 4. Hold a séance
codenecromancer seance examples/zombie_microservice

# 5. Perform the resurrection ritual
codenecromancer resurrect examples/zombie_microservice

# 6. Launch the Web Portal
codenecromancer serve --port 8000

Visit http://localhost:8000 to interact with the visual necrology lab!

🔗 Code Repository & MLH Project

GitHub Repository: https://github.com/FM9447/CodeNecromancer
MLH Project ID: 01a18aa2-6370-11ee-e4ba-3959179ae637

Built with 🖤 for Hacktoberfest 2026: Week 4.

SlothLink: The Lazy Developer's Autonomous Build-in-Public Agent

Faheem Musthafa P — Mon, 19 Oct 2026 18:11:54 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge: Week 3.

🦥 What I Built: SlothLink

October 20 is International Sloth Day, and sloths have life figured out: they move at their own rhythm, conserve every joule of energy, and let the world come to them.

As open-source developers, we love shipping features, squash-merging commits, and hacking on novel architectures. But what do most of us dread the most? Self-promotion and writing LinkedIn / build-in-public updates. It's tedious, mentally draining, and constantly interrupts the flow state.

Meet SlothLink: the lazy developer's autonomous build-in-public agent.

SlothLink runs quietly in the background on your machine. It monitors your GitHub repositories for new commits, watches your hackathon Gmail threads, performs cross-source correlation, and drafts authentic, feed-stopping technical posts. Best of all: it runs 100% locally using open-weight models (Gemma 4 and Qwen 2.5:3B) with zero cloud API token costs and built-in Deepwrite humanization.

📺 Live Studio Interface & Architecture

SlothLink is structured around an ultra-responsive, authentic LinkedIn desktop studio:

Left Column: Live telemetry cockpit, automated commit listeners, and active hackathon radar.
Center Canvas: Feed-stopping post recommendation cards, multi-turn AI chat editor, and 1-click publishing.
Right Column: Real-time token conservation tracker, local Ollama engine telemetry, and deepwrite parameter tuning.

🚀 Key Features

⏱️ Background Event-Driven Git & Mail Watchers:
- Actively polls local git branches and remote GitHub repositories (https://github.com/FM9447/linkdinsloth).
- Automatically triages commits: micro-refactors get grouped, while major architectural milestones trigger proactive draft alerts.
- Monitors hackathon acceptance emails and judging milestones to automatically correlate your repo code with event prize tracks.
🧠 100% Offline Inference (Best Use of Gemma & Qwen):
- Powered by local open-weight models via Ollama (gemma4:e4b and qwen2.5:3b).
- Incurs $0.00 cloud API credits and works completely offline without leaking proprietary source code or private emails to external SaaS providers.
✍️ Deepwrite Engine (0ms Fast De-AI + Local Style Re-synthesis):
- Strips corporate fluff, cringe LinkedIn jargon ("thrilled to announce", "game changer", "testament"), and robotic bullet lists.
- Uses zero-shot prompt framing to enforce authentic developer voice, code-diff citations, and natural storytelling.
🎯 Curated Post Archetypes:
- 🚀 The Milestone / Product Launch: Highlighting problem, solution, architecture, and live demo.
- 💡 Technical Breakdown & Architecture: Deep dive into local models, latency stats, and engineering tradeoffs.
- 🏆 Hackathon Demo Day: Formatted specifically for hackathon judges, highlighting sponsor tags and prize categories.
🔍 Cross-Source Correlation & Sequel Awareness:
- Understands your previous posts to avoid repetitive spam and synthesizes genuine sequel progress updates ("Part 2: After 30 days of testing...").

🛠️ Open-Source Tech Stack

Local AI Inference: Ollama running gemma4:e4b and qwen2.5:3b
Frontend: React 18, Vite, Tailwind CSS, Lucide React
Backend: Node.js, Express, simple-git, googleapis / imap-simple
MLH Project: SlothLink (Project #01a17457)

💡 Why Open Innovation Matters

Building in public shouldn't require trading your privacy or spending monthly subscriptions on cloud LLMs just to post an update about code you already wrote. By leveraging open-weight models like Gemma 4 running locally via Ollama, SlothLink proves that personal AI developer tools can be completely free, private, autonomous, and energy-conserving.

Work less, build more, and let your sloth handle the social media hustle.

🔗 Open-Source Repository

GitHub Repository: https://github.com/FM9447/linkdinsloth
MLH Project ID: 01a17457-a281-f087-f7e3-c95518f75009

Built with ❤️ for Hacktoberfest 2026: Week 3 (Slow and Steady).

Spot-Undoo: Privacy-First AI Kerala Micro-Travel Planner & Layover Explorer

Faheem Musthafa P — Sat, 17 Oct 2026 18:28:45 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge: Week 2 - Side Hustle.

📺 Watch the 30-Second Demo Video

👉 Watch Spot-Undoo in Action (Interactive Demo)

🌟 What I Built

Spot-Undoo is an open-source, privacy-first micro-travel companion and layover optimizer engineered for travelers across all 14 Kerala districts.

Most travel applications (Google Maps, TripAdvisor) funnel millions of tourists toward the same commercial bottlenecks while ignoring the quiet backwater channels, century-old bakeries, and heritage alleyways that embody the true soul of God's Own Country. Transit commuters stuck on railway layovers (like Shoranur Junction or Ernakulam South) suffer from extreme train delay anxiety and cannot find curated routes within strict time and rupee limits.

Spot-Undoo solves this by pairing locally hosted open-weight AI models (fine-tuned Qwen 2.5:3B / Qwen 3.5 4B) with MongoDB Atlas, ElevenLabs Voice AI, Vercel & Render cloud deployment, a Google Maps link & GPS parser, a Multimodal Ride Hailing Comparator (Uber, Rapido, Ola), Native Android Google Sign-In, 60fps OpenStreetMap with Avada Styles, 100m Proximity Guard with Interactive Place Previews, and a Native Android APK to generate constraint-guaranteed itineraries on the fly.

📸 Latest App Screenshots

🚀 Key Features

🗺️ 8 Curated Kerala Hubs & 75+ Verified Spots:
- Comprehensive authentic spots across Shoranur Jn, Pattambi, Kochi / Ernakulam, Calicut, Palakkad, Trivandrum, Kollam, and Wayanad.
- Curated high-resolution photography with automatic onerror CDN fallbacks so no spot ever shows broken icons or mismatched waterfalls.
🛺 Multimodal Ride Hailing Comparator (Uber, Rapido, Ola):
- Live fare calculation engine based on Kerala Transport meter rates and distance.
- Side-by-side comparison for Rapido Bike Taxi, Uber Moto, Uber Auto, Rapido Auto, Ola Auto, and Uber Go (AC Cab).
- Smart ranking filters: Lowest Fare, Fastest ETA, Bikes, Autos, Cabs.
- 1-tap "Go" button directly opens the native Uber or Rapido mobile app with the destination coordinates and place name pre-filled!
📍 Smart Location Engine & 100m Proximity Guard:
- Paste any Google Maps URL (including short links like maps.app.goo.gl, @lat,lng, !3d!4d, or query=lat,lng) to automatically extract coordinates, place names, and detect the nearest Kerala city.
- Interactive 100-Meter Proximity Verification: Detects when a spot already exists within 100 meters, opening an interactive confirmation modal showing the existing place's photos, rating, and reviews.
- One-tap live GPS auto-fill with instant local cache fallback.
🧠 Edge AI Itinerary Synthesis & Zero-Duplicate Engine:
- Local on-device Qwen inference on consumer GPUs (71 tokens/sec) with zero per-token cloud costs.
- Generates 3 distinct JSON itineraries: Plan 1 (Mood Match), Plan 2 (Ultra-Budget Saver), and Plan 3 (Hidden Gem Route).
- Strict Zero-Duplicate & Termination Sanitizer: Guarantees every single stop is unique; never adds repetitive loops or duplicate return legs to fill time.
🗺️ 60fps OpenStreetMap with Avada Themes:
- Silky smooth 60fps pan and zoom with dynamic multi-level zoom photo markers and viewport culling.
- Instant map theme switching between Standard OSM, Dark Mode, High-Contrast Neo-Brutalist, and Voyager.
📱 Native Android Google Sign-In & APK:
- Powered by @codetrix-studio/capacitor-google-auth for seamless one-tap native account selection on Android without opening external browser tabs.
- Packaged with Capacitor for native Android deployment (SpotUndoo-v1.0.apk).
🎙️ ElevenLabs Multilingual Audio Tour Guide:
- Built-in audio narration for spots and transit trails powered by ElevenLabs with automatic client-side Web Speech fallback.
- Travelers can listen to spot history and insider tips hands-free while walking.
🍃 MongoDB Atlas Cloud Integration & Firebase Storage:
- Multi-user authentication system and user-specific itinerary history.
- Cloud persistence for community-uploaded spots and traveler reviews with Firebase photo storage.

🛠️ Tech Stack & Architecture

Frontend & CDN: React 18, Vite, Tailwind CSS, Lucide Icons, Leaflet OpenStreetMap, hosted on Vercel Edge Network.
AI Model & Inference: Local Qwen 2.5:3B on Ollama (spot-undo-qwen) + Groq Cloud API (Qwen 3.8 27B) + Google Gemini Flash + Tinker SDK fine-tuning.
Ride Hailing Integration: Deep linking into Uber (uber://), Rapido (rapido://), and Ola (olacabs://) with dynamic fare calculation.
Audio & Voice: ElevenLabs Multilingual TTS (/api/audio/tts) with Web Speech API fallback.
Database & Cloud Backend: MongoDB Atlas cloud cluster + Firebase Storage + Node.js / Express REST API deployed on Render.
Auth: Native Android Google Sign-In (@codetrix-studio/capacitor-google-auth), Firebase Auth, and bcrypt JWT.
Mobile Engine: Capacitor native Android bridge (SpotUndoo-v1.0.apk).

🔗 Code Repository & Links

Interactive Demo Video: Watch on ngram
GitHub Repository: https://github.com/FM9447/Spot-Undoo
Slide Deck (.pptx): Spot-Undoo Presentation Deck
MLH Project: MLH Project #01a14abe

Built with ❤️ for Hacktoberfest 2026 & MLH Hackathons.