Results: 34% returns over 3 months vs 7% S&P 500. Real capital, not backtesting.
Architecture:
Stage 1: LightGBM ranks 1,700+ stocks daily (80+ features) Stage 2: JAX PPO optimizes portfolio allocation Walk-forward validation, champion/challenger deployment Nightly retraining pipeline Recent additions:
Personal portfolio analysis with ML health scores Daily digest with Perplexity-summarized news Market regime detection (bull/bear + volatility) Stack: FastAPI, React, PostgreSQL, LightGBM, JAX/Flax, Claude API
Now seeking acquisition by brokerages, robo-advisors, or fintech platforms. Building this from scratch is 2-3 years of specialized quant work.
Live demo: https://acis-trading.com
Happy to discuss the ML architecture, validation methodology, or acquisition strategy.