RExplain 🔍

AI-Powered GitHub Repository Analysis Platform

🚀 Live Demo → rexplain.vercel.app

RExplain automatically explains any GitHub repository using static analysis, architecture extraction, and Retrieval-Augmented Generation (RAG) — making it dramatically faster to understand large codebases without reading every file manually.

Simply paste a GitHub URL and get an instant, AI-powered breakdown of the codebase.

---

✨ What It Does

Feature	Description
🧠 Framework Detection	Detects FastAPI, Flask, Django, React, Vue, Angular, SQLAlchemy, PostgreSQL, MongoDB, and more
🏗️ Architecture Diagrams	Generates dependency graphs and visual architecture diagrams
📄 README Extraction	Fetches and renders the repository README with syntax-highlighted code blocks
📊 Commit Analytics	Visualizes commit history, frequency, and contribution graphs
🤖 AI Chatbot (RAG)	Ask natural-language questions about any repository
⚡ Instant Reloads	PostgreSQL-backed cache skips re-analysis on unchanged repositories

---

🧱 Tech Stack

Backend

• FastAPI — REST API framework
• PostgreSQL + SQLAlchemy — persistent storage and ORM
• SentenceTransformers (all-MiniLM-L6-v2) — embedding generation
• Groq API — LLM inference
• Graphviz — architecture diagram rendering
• GitHub REST API — repository metadata and tree fetching
• Docker — containerized deployment

Frontend

• React + Vite — fast frontend build
• TailwindCSS — utility-first styling
• Framer Motion — animations
• React Markdown — README rendering
• Axios — HTTP client

---

🚀 How It Works

GitHub URL
   ↓
Repository metadata extraction
   ↓
Selective file fetching (GitHub Tree API)
   ↓
Framework detection
   ↓
Architecture generation
   ↓
README + intelligence extraction
   ↓
RAG indexing
   ↓
Interactive AI chat

---

🔑 Key Engineering Decisions

GitHub Tree API + Selective Fetching

Instead of cloning entire repositories (which causes bottlenecks for large repos, binaries, node_modules, and datasets), RExplain:

• Fetches the repository tree structure via the GitHub API
• Identifies and selectively retrieves only important files — README, package.json, requirements.txt, route files, configs, and manifests
• Falls back to a shallow clone (depth=1) only if the API is unavailable

Result: Fast analysis regardless of repository size, with no large clone delays.

Persistent Embedding Cache

Embeddings and chunk metadata are stored in PostgreSQL so repeated loads are instant:

1. Fetch the latest commit SHA and compare with the cached SHA
2. If unchanged → restore embeddings and chunks, skip full analysis
3. If changed → run the full pipeline and persist the new state

Startup Model Loading

The embedding model loads once at backend startup rather than per-request, eliminating timeout issues during AI chat.

---

🤖 AI Chatbot (RAG)

Users can ask natural-language questions about any analyzed repository:

• "What frameworks are used?"
• "Where is the database set up?"
• "How does authentication work?"
• "What are the main API routes?"

How it works:

1. Repository files are chunked and embedded using all-MiniLM-L6-v2
2. On each question, relevant chunks are retrieved via semantic similarity search
3. Retrieved context is passed to an LLM (Groq) for a grounded, repository-specific answer

---

🏗️ Architecture

┌─────────────────────────────────┐
│        React Frontend           │  ← Vercel
│  (Analysis Panel + AI Chat)     │
└────────────────┬────────────────┘
                 │ HTTPS
┌────────────────▼────────────────┐
│        FastAPI Backend          │  ← Render (Docker)
│  Analysis · RAG · Cache Layer   │
└──────┬──────────────────┬───────┘
       │                  │
┌──────▼──────┐   ┌───────▼──────┐
│  PostgreSQL  │   │  GitHub API  │
│  (Neon)    │   │  Groq API    │
└─────────────┘   └──────────────┘

---

☁️ Deployment

Layer	Platform
Frontend	Vercel
Backend	Render (Docker)
Database	Neon PostgreSQL

Docker is used for the backend because Graphviz requires system-level installation, which Render's standard Python runtime does not support natively.

---

⚠️ Known Limitations

• GitHub API Timeouts — Very large repositories may be slower to analyze; retry logic and graceful fallbacks are in place
• Cold Starts — Render's free tier sleeps inactive services; the first request after inactivity may take a few seconds longer

---

🗺️ Roadmap

• Interactive Diagrams — Replace Graphviz with React Flow for clickable, interactive dependency graphs
• Advanced RAG — Hybrid retrieval, AST-aware chunking, reranking, and source citations
• Streaming Responses — Real-time streaming AI chat
• Multi-Repository Analysis — Analyze microservice ecosystems and org-wide dependency relationships
• Background Workers — Celery + Redis for async job pipelines

---

🚀 Try it live at rexplain.vercel.app