A full-stack application for practicing stock trading with ₹10,00,000 of virtual capital, extended with an AI Research feature — upload financial PDFs and ask grounded questions about them (RAG).
- Live simulated stock prices that update every 3–5 seconds
- Buy/sell with a wallet (₹10,00,000 starting balance)
- Real-time portfolio P&L tracking
- Full transaction history with filters
- Dashboard watchlist with auto-refreshing prices
- JWT authentication with bcrypt password hashing
- Upload a 10-K, annual report, or earnings PDF
- Ask natural-language questions about it
- Answers cite the exact source passages used
- Download the original source PDF to verify LLM answers against the source
- Backed by Gemini embeddings + pgvector similarity search
- A machine-learning model predicts whether a stock closes up or down the next day
- Trained in Python (scikit-learn) on technical-indicator features, with a documented Jupyter notebook (EDA → train/test → evaluation → feature importance)
- Served by a FastAPI microservice; shown in the UI with probability, confidence, and the top contributing signals
- See ml-service/README.md for the model card and interview notes
| Layer | Technology |
|---|---|
| Frontend | React 19 + Vite, Tailwind CSS, React Router, Recharts, Three.js, Lucide icons |
| Backend | Node.js + Express.js, JWT auth, bcryptjs |
| Database | PostgreSQL with Prisma ORM + pgvector extension |
| AI | Google Gemini (gemini-2.5-flash + gemini-embedding-001), pdf-parse, vanilla-JS RAG |
| ML | Python · FastAPI · scikit-learn · pandas (price-prediction microservice + Jupyter notebook) |
| Deployment | Frontend on Vercel, backend on Render, database on Neon |
stock-project/
├── backend/ Express API (trading, portfolio, AI)
│ ├── ai/ RAG pipeline (PDF → chunks → embeddings → answers)
│ ├── controllers/ HTTP handlers
│ ├── middleware/ JWT, error handler, multer upload
│ ├── prisma/ Schema + migrations (includes pgvector)
│ ├── routes/
│ ├── scripts/ Seed scripts (AI samples + test user portfolio)
│ └── services/
├── frontend/ React SPA
│ └── src/{pages,components,services}
├── ml-service/ Python ML microservice (FastAPI + scikit-learn)
│ ├── train.py training + evaluation pipeline
│ ├── app.py FastAPI serving layer
│ └── notebooks/ Jupyter notebook (full ML walk-through)
├── RAG.md In-depth architecture doc for the AI feature
├── render.yaml Backend + Postgres deploy config (Render)
├── vercel.json Frontend deploy config (Vercel)
└── README.md
- Node.js 18+
- A PostgreSQL database with the
pgvectorextension. The easiest option is a free Neon project — pgvector is available on free tier.
git clone <repository-url>cd stock-projectcd backend
npm install
cp .env.example .env # then fill in DATABASE_URL, JWT_SECRET, GEMINI_API_KEY
npx prisma generate
npx prisma migrate deploy # runs both migrations, enables pgvector
npm run dev # starts on http://localhost:3000cd frontend
npm install
npm run dev # starts on http://localhost:5173cd ml-service
python3 -m venv .venv &&source .venv/bin/activate
pip install -r requirements.txt
uvicorn app:app --port 8000 # serves price-direction predictionsThe trained model is committed, so it runs immediately. See ml-service/README.md to retrain or use real market data.
cd backend
npm run seed:ai # seeds 3 sample financial reports for every usercd backend
node scripts/seedTestUserPortfolio.jsResets testuser@gmail.com to a 60-day trading history (13 trades, 9 holdings).
The fastest way in: click "🚀 View Live Demo" on the landing page (or "Try Demo Account" on the login page) to log straight into a pre-populated account — no typing needed.
| Account | Password | |
|---|---|---|
| Demo | demo@demo.com | demo123 |
| Test | testuser@gmail.com | test123 |
To (re)create the demo account's data:
cd backend
npm run seed:demo # holdings, transactions, wallet
node scripts/seedSamples.js demo@demo.com # 3 sample AI Research documents| Service | Port | Purpose |
|---|---|---|
| Frontend (Vite) | 5173 | React app |
| Backend (Express) | 3000 | API + JWT auth + AI orchestration |
| ML service (FastAPI) | 8000 | Stock price-direction predictions (scikit-learn) |
| Database (Postgres) | 5432 | Trading data + AI document chunks + vectors |
In dev, the React app uses Vite's proxy to talk to localhost:3000.
# Run frontend + backend together (from root)
npm run dev
# Stop everything
pkill -f "nodemon server.js"
pkill -f "vite"# Reseed AI samples for one usercd backend && npm run seed:ai user@example.com- Frontend — Vercel (SPA rewrite via
vercel.json) - Backend — Render (config in
render.yaml) - Database — Neon (PostgreSQL with pgvector)
For deployment, set the same env vars from .env.example in your hosting platform. Run prisma migrate deploy against the production database once.
- RAG.md — full architecture and interview prep notes for the AI feature
Created with ❤️ for Advanced Agentic Coding by Ansh Baheti.