Volatility Harvesting Engine

Overview

VHE is a single-process algorithmic trading terminal for Indian NSE equities that runs autonomously after start-up. It ingests live quotes (yfinance, a simulated tick generator, or the paid Zerodha Kite WebSocket), computes EMA/ATR/ADX indicators on every tick, classifies each symbol's market regime as RANGE, TREND, or CRASH, and runs three capital-allocated strategies through a shared risk engine before routing orders to a paper or live broker. A FastAPI backend streams full platform state over WebSocket to a React/TypeScript dashboard behind Google OAuth login, backed by a Monte Carlo risk engine and a walk-forward validation harness to guard against overfitting. It's deployed 24/7 on an Oracle Cloud Always Free ARM instance behind nginx and systemd, with 147 passing tests.

Architecture

1. 01Quote feed (yfinance, simulated generator, or Zerodha Kite WS) ticks every 30s into a shared PlatformRuntime.
2. 02IndicatorService and RegimeService update EMA-20/ATR-14/ADX-14 and classify each symbol as RANGE, TREND, or CRASH.
3. 03StrategyOrchestrator runs the DynamicGrid, PairSpread, and Momentum strategies against the current regime and capital allocation.
4. 04RiskGuard checks every order against loss, exposure, and sentiment rules before ExecutionEngine routes it to PaperBroker or KiteBroker.
5. 05State is persisted to SQLite and broadcast over WebSocket to all connected dashboard sessions.

What I learned

• Building a regime-aware strategy orchestrator that arms and disarms strategies based on live ADX and drawdown signals.
• Using Monte Carlo bootstrapping and walk-forward splits to validate that a trading strategy isn't just overfit to history.