RepoFlow

A GitHub-style developer platform with its own version-control CLI, issue tracking with AI duplicate detection, realtime notifications, and peer-to-peer code review calls.

Built end-to-end (React frontend + Node backend + AI + cloud infra) from a single repo.

---

Features

• Custom Git-like CLI — login, init, add, commit, push, pull, revert from any local folder; commits sync to RepoFlow's backend with a local .repoFlowGit working folder.
• Repositories — create public/private repos, star/unstar, explore top-level repo listing, and browse commit history with file-level revert.
• AI duplicate issue detection (RAG) — new issues are embedded (Google Gemini embeddings) and matched via semantic search (Pinecone, per-repo namespaces); the top candidates then ground a generative Gemini verdict that flags duplicates and explains why. Debounced 600ms on the frontend + 5-min result caching to keep costs in check.
• Realtime notifications — Socket.IO pushes events live; every notification is also persisted in MongoDB so nothing is missed while offline.
• Code review rooms — 1-on-1 WebRTC audio/video calls launched from a commit, with the changed files displayed side-by-side. Socket.IO handles only signaling; media flows peer-to-peer via STUN/TURN.
• Profiles & contribution heatmap — GitHub-style activity graph, starred-repos tab, and per-action contribution logging (repo created, starred, issue opened/closed, etc.).
• Fault-tolerant push — commit files go to AWS S3 with an automatic MongoDB base64 fallback (storageType), so pushes work even when object storage is down.

---

Tech Stack

Layer	What's used
Frontend	React 19 + TypeScript · Vite · Tailwind CSS · React Router 7 · Axios · Socket.IO client · WebRTC (RTCPeerConnection)
Backend	Node.js · Express 5 · MongoDB + Mongoose · JWT auth (bcrypt + middleware ownership checks) · Socket.IO · Redis cache · AWS S3
AI	Google Gemini embeddings (768-dim) · Gemini generative verdicts · Pinecone vector store
Infra	GitHub Actions CI · AWS Amplify (frontend deploy) · EC2 + PM2 cluster (backend) · Docker Compose (local Redis)

Architecture at a glance

React SPA (Vite, TypeScript)
       │  REST (Axios, JWT) + Socket.IO + WebRTC
       ▼
Express API + Socket.IO server
   ├── MongoDB …… users, repos, issues, commits, notifications, contributions
   ├── AWS S3 ……. commit files (falls back to MongoDB base64 on failure)
   ├── Redis ……... API caching (bounded retry/back-off — never takes the app down)
   └── Pinecone …. per-repo issue vector namespaces for AI search

---

Design Decisions

• Commit storage with graceful degradation — try S3 first, fall back to MongoDB if unavailable; the read path picks the source via storageType, so clients never see a failed push.
• Semantic duplicate detection — embeddings beat keyword matching for issues phrased differently but meaning the same thing; results are isolated per repository.
• Redis as a best-effort layer — no REDIS_URL, unreachable Redis, or timeouts just degrade reads to MongoDB with a 30s back-off instead of failing requests.
• Notifications that survive offline moments — write to MongoDB and emit over Socket.IO in one go, so the unread count is always correct.
• Review calls without media in the server — Socket.IO does room signaling (max 2 participants); the actual audio/video is true peer-to-peer via WebRTC with STUN/TURN.

---

Project Structure

RepoFlow/
├── Backend/                  # Express API + custom version-control CLI
│   ├── controllers/          # auth, repos, issues, AI, CLI, commits, notifications
│   ├── helpers/              # redis cache, socket.io, vector store, embeddings, signaling
│   ├── Middleware/           # JWT auth + ownership authorization
│   ├── model/                # Mongoose models
│   ├── routes/               # REST routers
│   └── tests/                # Jest + supertest + mongodb-memory-server
├── frontend/                 # React + TypeScript SPA (Vite)
│   ├── src/components/       # auth, dashboard, repo, issues, review, user pages
│   └── src/config/           # axios, socket, webrtc clients
├── .github/workflows/ci.yml  # CI: backend tests + frontend build
├── docker-compose.yml        # local Redis
└── amplify.yml               # AWS Amplify frontend deployment

---

Run Locally

Prereqs: Node.js 20+, MongoDB, Redis (optional), and a Backend/.env with your MongoDB URI, AWS keys, PINECONE_API_KEY/PINECONE_INDEX, and GOOGLE_API_KEY.

# Backend API  →  http://localhost:3000
cd Backend
npm install
npm start

# Frontend SPA  →  Vite dev server (default http://localhost:5173)
cd frontend
npm install
npm run dev

Optional local cache:

docker compose up -d redis   # Redis on :6379 (production uses hosted REDIS_URL)

CLI usage

cd Backend && node index.js login --email you@example.com --password "***"
node index.js init my-repo
node index.js add file.js
node index.js commit "Initial commit"
node index.js push

---

Testing & CI/CD

• Backend tests — Jest + supertest against an in-memory MongoDB (mongodb-memory-server); suites cover auth, CLI init/push, and repo flows. No external services needed: cd Backend && npm test.
• GitHub Actions — runs backend tests and builds the frontend on every push/PR to main.
• Deployment — frontend auto-builds and deploys on AWS Amplify; backend runs as a PM2 cluster (one worker per CPU core) on EC2 with auto-restart.

---

RepoFlow is a self-driven, portfolio-level project: no boilerplate, real infra, real AI, real tests.