Ghost Mentor Match：結合 Qwen 與 Gemma 的 AI 編程導師配對及程式碼追思平台

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

Every software project is haunted. Beneath the clean production UI lives a graveyard of deleted functions, abandoned architectures, TODO comments from years ago, and 2:00 AM debugging sessions where the machine seems to have a mind of its own.

Inspired by the OctoberFest mentor-match architecture and this week's Ghost in the Machine theme, I built Ghost Mentor Match: The Spirit in the Machine—an open-source AI mentorship and Día de los Muertos code-remembrance platform.

The application bridges both halves of the challenge prompt through two interactive experiences:

1. The Séance Mentor Match (Haunted Debugging & Architecture): Instead of generic chatbot answers, developers input their current tech stack, stack trace, or architectural roadblock. The system calculates compatibility scores and matches them with specialized "Coding Spirits"—AI mentor personas channeling the wisdom of Unix pioneers, systems kernel phantoms, and distributed-systems veterans. You open a live "Séance Session" where the mentor diagnoses your bug with eerie terminal poetry backed by concrete, senior-level engineering fixes.
2. Día de los Muertos Digital Code Ofrenda: In the spirit of November 1 and 2, developers can paste deleted git log commits, deprecated modules, or sunsetted side projects onto a glowing digital altar adorned with marigolds (cempasúchil) and candlelight. The open-weight model reads the diff context and composes a heartfelt tribute honoring how that deleted code gave its life so today's production build could survive.

Screenshots

1. Ghost Mentor Matching & Compatibility Dashboard

1. Live Séance Debugging Chat (Powered by Qwen 3.8-27B & Local Gemma)

Code

🔗 GitHub Repository: https://github.com/CodeWithPritom/mentor-match

How I Built It

I designed and built Ghost Mentor Match with an open-source AI core from the ground up:

• Open-Weight Models at the Core (Qwen 3.8-27B & Google Gemma 2): The primary reasoning and persona-matching engine is powered by the open-weight Qwen 3.8-27B (qwen-3.8-27b) model running on ultra-fast Cerebras Cloud inference, paired with a native local inference fallback using Google's Gemma (gemma2:2b) via Ollama (http://localhost:11434).
• UI/UX Design in Figma Make: I crafted the dark obsidian interface (#090A0F) in Figma Make, blending spectral cyan terminal diagnostics (#00F5D4) with warm Día de los Muertos marigold gold (#FFB703).
• Zero-Lock-In Architecture: Built on top of the OctoberFest mentor-match workflow using modern React/Tailwind and an OpenAI-compatible open-weight inference adapter. Switching between cloud-hosted open weights (Cerebras) and 100% offline local weights (Ollama + Gemma) takes a single toggle in the UI.

Why Does Open Innovation Matter?

Building a developer mentorship tool that inspects broken stack traces and private git histories highlights exactly why open-weight models and open innovation beat closed, proprietary APIs:

1. Total Privacy for Proprietary Code & Git History: When you are debugging a broken authentication flow or building an ofrenda from your repository's git log, you are handling internal file paths, business logic, and commit histories. Running Gemma locally or routing through self-controlled open weights ensures sensitive source code isn't harvested to train a closed vendor's next commercial model.
2. Works When the Lights (and Internet) Go Out: A true "Ghost in the Machine" shouldn't crash with an APIConnectionError when your Wi-Fi drops during a late-night coding session. By supporting local open weights on consumer hardware, developers can summon mentor diagnostics completely offline at $0.00 token cost.
3. No Disappearing Spirits (Zero Vendor Lock-In): Closed APIs deprecate model checkpoints overnight, breaking carefully tuned system personas. Because Qwen and Gemma are open-weight models, the exact weights powering our mentor personas and Día de los Muertos tributes belong to the community forever.

My Agent Session

During the rapid build sprint from the OctoberFest makeathon to final deployment, I used Cursor alongside GitHub Copilot to scaffold the multi-model inference adapter (bridging Cerebras qwen-3.8-27b and local Ollama gemma2:2b), wire up the interactive Día de los Muertos altar state, and refine the responsive Figma Make UI components.

The Spirit of It

The moment this project truly landed was when I ran a real commit log from an old, abandoned student project through the Día de los Muertos Code Ofrenda.

Seeing a deleted 300-line authentication module honored on screen—with glowing marigold cards reminding me that "no code is truly wasted if it taught the hands that wrote the next version"—genuinely hit different. As developers, we spend so much time deleting, refactoring, and moving on that we rarely pause to appreciate the messy early commits that carried us forward. Turning git log from a dry audit trail into a living altar of engineering memories captured the exact thin line between past and present that Halloween and Día de los Muertos represent.