Personal project · Sole developer · Live as Prajna
AutoTrade Pro
Autonomous AI Trading Platform
An autonomous system that ingests Indian equity market data, scores the full NSE universe on a schedule, and executes trades through a layered risk engine.
Key metrics
- lines of Python/FastAPI backend
- 96,000
- lines of React frontend
- 33,000
- PostgreSQL tables
- 55
- scheduled jobs
- 40+
What it does
- 26 REST services, 55 PostgreSQL tables and 37 dashboard pages behind one platform.
- Event-driven pipeline on Celery and Redis running 40+ scheduled jobs, from 5-second stop-loss sweeps to 15-minute NSE-wide scoring cycles.
- Multi-factor scoring engine fusing technical, news, macro, earnings and options signals; every decision snapshot persisted for audit and backtest replay.
- LLM agent layer on AWS Bedrock (Amazon Nova Pro) with tool calling for event classification and trade-reasoning gates, plus streaming SSE chat over the portfolio.
- Per-symbol LSTM classifiers (TensorFlow/Keras) and scikit-learn ensembles, time-series cross-validated.
- NLP news subsystem: crawlers across 7+ sources, sentiment scoring, and an entity-impact graph for second-order candidates, with LLM parsing of earnings-call PDFs.
- Risk engine with drawdown circuit breakers, consecutive-loss lockouts, correlation-cluster guards, sector-exposure caps and volatility-aware sizing.
- Live broker integration via Zerodha Kite and Upstox — WebSocket tick streams, TOTP auth, token lifecycle and order execution.
- Deployed on Linux with systemd services and Telegram alerting.
Stack
- Python
- FastAPI
- Celery
- Redis
- PostgreSQL
- React
- TensorFlow
- scikit-learn
- AWS Bedrock
- Docker