A multi-agent trading system where LLM proposers discover opportunities, adversarial debate agents stress-test every thesis, and a risk manager gates execution. Proposers compete in a credit-based tournament against rule-based baselines.
Scans entire US equity universe every 5 minutes. Cross-references price action with news catalysts. Ranks signals by composite score.
Every trade proposal faces a bull and bear agent. The risk manager synthesizes both sides blind — it doesn't know who proposed.
Shadow P&L tracks every rejected proposal. The system learns which rejections were smart and which were missed opportunities.
iOS app that uses LLM sampling and reasoning to intelligently schedule your day. Protects deep work, respects constraints.
Zero frameworks, static HTML/CSS/JS. Served by Caddy with automatic HTTPS via Let's Encrypt. Hosted on DigitalOcean NYC1.
Make it work, make it correct, make it fast. In that order. A running system that generates data is worth more than a perfect system that's never deployed.
Every trade, every rejection, every market regime shift is training data. Build feedback loops that turn failures into calibration, not just log entries.
Good ideas survive scrutiny. Every proposal gets a bull case, a bear case, and a risk review before it touches real capital. Confirmation bias is the enemy.
Humans should make judgment calls, not babysit cron jobs. If a machine can do it reliably at 3 AM, it shouldn't require a human at 9 AM.