📌 Pinned A self-improving agentic quant research harness with a human in the loop

A self-improving multi-agent coding system. Agents implement each hypothesis independently, review each other's code, reason about what's holding a strategy back, and test alternative fixes, while a registry of every past experiment keeps failed ideas from returning. Correctness comes from forced agreement, AI-assisted human verification across market regimes, multiple independent backtest engines, and a seven-gate path to trust. I set the direction, the agents automate the search, and the combination finds better paths than either of us would alone.

Read more

📌 Pinned An agent-friendly charting tool for inspecting backtests

A flexible, open-source charting tool I built for my quant research. It renders candles, trades, stops, and indicators in one HTML file, zooms like a paid trading platform, and can be driven by AI agents that add overlays, pick the window, and label what to review.

Read more Open source