DEV Community

Cover image for RepoGraveyard 3D: Resurrecting Legacy Code with Gemma 4
Akul
Akul

Posted on

RepoGraveyard 3D: Resurrecting Legacy Code with Gemma 4

Hacktoberfest Open-Source AI Challenge: Ghost in the Machine Submission 👻🌼

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine

What I Built

RepoGraveyard 3D is an atmospheric software archaeology workstation built for the "Ghost in the Machine" challenge. Every forgotten open-source repository carries a ghost—abandoned TODOs, breaking dependency changes, and bit-rotted toolchains.

RepoGraveyard 3D identifies and repairs legacy open-source bit-rot using local Gemma 4 9B AI orchestrated inside an interactive retro 3D WebGL CRT server room.

Demo

Live Interactive Workstation: https://banxal.tech

Code

GitHub Repository: https://github.com/Akulbanxal/graveyard-3D

How I Built It

  1. 3D WebGL Terminal: Built with React Three Fiber, Three.js, and custom post-processing GLSL glitch shaders.

  2. Local AI Inference: Local Gemma 4 9B model (via Ollama) executing stack-trace diagnostics, toolchain failure analysis, and unified git diff generation.

  3. Agent Orchestration Framework: Managed using the Mastra Agent Framework.

  4. Observability & Tracing: Real-time agent tracing, token counts, and span latencies instrumented via @sentry/nextjs.

  5. Infrastructure: Production Web Service deployed on Render bound to our custom .tech domain (banxal.tech).

  6. Narration: ElevenLabs synthesized ghost voiceover for screen walkthrough audio.

Why Does Open Innovation Matter?

Open-weight models like Gemma 4 make private, local software archaeology possible. Legacy codebases and commit histories often contain sensitive security logs, internal structural data, or personal details that developers cannot share with proprietary cloud APIs. Running Gemma 4 locally via Ollama guarantees data sovereignty, zero API costs, and full offline resilience while granting full control over system prompts and agent execution parameters.

My Agent Session

This application was designed and iterated alongside agentic AI coding assistants. Below is an authentic execution trace payload generated during local Gemma 4 code exhumation: {
"target_repo": "octocat/hello-world",
"status": "RESURRECTED",
"agent_framework": "Mastra v1.0",
"model": "gemma-4-9b-instruct",
"telemetry": {
"sentry_trace_id": "8f3b21a09c4d4e12",
"execution_latency_ms": 1420
},
"diagnostics": {
"bit_rot_cause": "Deprecated peer dependency resolution manifest",
"patch_status": "Clean git diff synthesized"
}
}

The Spirit of It

Software archaeology is about remembering the human story behind the code. Stepping into the 3D server room with flickering CRT monitors, low ectoplasm fog, and the low hum of local AI inference transforms routine debugging into an atmospheric experience—giving forgotten open-source projects a second life.

Prize Categories

  1. Overall Winner: "Ghost in the Machine" interactive 3D WebGL workstation implementation.
  2. Best Use of Gemma: Local Gemma 4 9B model handling automated code repair and diagnosis.
  3. Best Use of Render: Production app runtime and Next.js frontend deployed live on Render.
  4. Best Use of Mastra: Workflow orchestration across agent execution steps.
  5. Best Use of Sentry Agent Tracing: Distributed tracing of model latency and agent spans.
  6. Best Use of a .tech Domain: Hosted live at banxal.tech.
  7. Best Use of GitHub Copilot: Built with Copilot CLI and GitHub Actions workflows.
  8. Best Use of ElevenLabs: Demo video voiceover narration.
  9. Best Use of Entire: Linked agent prompt session traces.

Top comments (0)