Talk to any codebase. Understand it in seconds.
GitScope is an 8-stage analysis engine that combines deterministic code intelligence with Gemini's 2M-token context to give you full visibility into any GitHub repository. Paste a repo, get a complete diagnostic profile, then chat with the codebase.
Built solo at hackUMBC Mini Hackathon 2026 (12 hours).
- 8-stage analysis pipeline — file ranking, secret scanning, dependency extraction, code metrics, all validated by Gemini AI
- Chat with any codebase — ask questions, get answers grounded in real files with line numbers
- Narrated walkthrough — 6-scene guided tour with TTS and data-driven visualizations
- Contributor onboarding — beginner-friendly file suggestions, setup steps, improvement ideas
- Head-to-head comparison — compare two repos side-by-side across health, security, and quality
- Interactive architecture graph — SVG node diagram with click-to-view source
GitHub API Fetch → Clone repo, parse file tree, extract source files File Importance Ranker → Score by import frequency, depth, size Secret Scanner → 12 regex patterns for API keys, passwords, credentials Dependency Extractor → Parse package.json, requirements.txt, etc. Code Metrics Engine → LOC, comment ratio, long files, tech debt ↓ Local findings + full codebase ↓ Gemini 1.5 Pro → Validate findings against real source code (2M context) ↓ Post-Processor → Merge local + AI findings, compute health score Interactive Dashboard → Chat, walkthrough, architecture graph, code viewer
Stages 1-5 are deterministic local analysis. Stage 6 is the only AI call. AI validates — it doesn't originate.
| Layer | Tech |
|---|---|
| Backend | Python / Flask |
| Frontend | React 18 (Vite) |
| AI Engine | Gemini 2.5 Flash |
| Context Window | 2M Tokens |
| Data Source | GitHub REST API |
| Static Analysis | 12 Regex Patterns |
| Visualization | Hand-written SVG |
| TTS | Web Speech API |
pip3 install flask flask-cors requests google-generativeai
export GEMINI_API_KEY='your-key-here'
export GITHUB_TOKEN='your-token-here'
python3 app.py
cd frontend && npm install && npm run devMIT