I build ATLAS, a multi-agent research system that looks for a tradeable edge in Nasdaq-100 futures — and, so far, honestly reports that it hasn't found one.
- 11 analysis agents and 5 pluggable strategies, trading autonomously on an Interactive Brokers paper account, unattended for weeks at a time
- A statistical validation harness — bootstrap confidence intervals, permutation tests, walk-forward — gates every strategy before real money is allowed near it
- Negative results are published, not hidden: profit factor ≈ 1.0 on directional OHLCV signals across 8 years of tick-derived data
Featured project: trading-agents — the full system: execution engine, agents, validation harness, and desktop / web coaching apps built on top of it.
Python · TensorFlow · scikit-learn · XGBoost · pandas · SQLite ·
Starlette · Interactive Brokers API · Databento