AURIGA Autonomous Quant Research&Investment Agent

AURIGA is an autonomous research agent that mines 5 years of 1H/1D Alpaca data to generate explainable options strategies, validates them statistically, and trades on a $100k Alpaca paper account—LLM has no execution authority. Three ML engines: A1 (Strategy): extracts LONG/SHORT rules from XGBoost tree paths (9 admitted rules: Sharpe ≥2, Win Rate ≥0.65, Profit Factor ≥1.5, ≥30 trades, BH/FDR-controlled). A2 (Vol Risk): predicts vol shocks (AUC≈0.75) as a risk signal, not a strategy—buying vol yields median Sharpe -2.6 (variance risk premium). A3 (Premium Selling): sells credit spreads while learning AVOID_SELL rules for high-risk regimes. Orders pass 8 fail-closed risk gates (daily loss, asset/sector/total limits, position caps, liquidation, A2 danger, A3 avoid-sell), executing atomically via Alpaca MLEG orders, monitored via Streamlit dashboard + daily LLM narratives. Backtests use a synthetic Black-Scholes proxy (free-tier data limits); final 20% holdout stays untouched.