Diamond Hands
Something I'd wanted to build for a long time: a multi-agent system that runs my own trading process. Strategy rules find trades in stocks and crypto, Claude agents vet them, and a risk engine sizes the orders.
Why I built it
I trade on the side, and for a long time my process lived in my head and a lot of manual checking: scanning for setups, deciding whether a trade was worth taking, and working out how much to risk. It took time, and it only worked if I followed my own rules every single time.
So I built the system I wanted, modelled on a small trading firm. Strategy rules are the analysts who spot opportunities, Claude agents are the review desk that questions each one, and a risk engine is the risk manager with the final say on size. It can run on autopilot, or send every trade to my phone on Telegram so nothing goes through until I approve it.
It's also become my playground for the things I enjoy most as a developer: multi-agent AI, testing ideas properly with data, and running small services reliably in the cloud.
Highlights
- Backtests every strategy bar by bar, with walk-forward and Monte Carlo tests, before it goes live.
- Strategy rules scan each market and propose trades, and Claude agents review every one before it goes any further.
- Runs on autopilot, or sends each trade to Telegram so I can approve it first.
- A risk engine sizes each order, caps total exposure and fails closed: if the data looks wrong, it doesn't trade.
- Each market runs as its own bot on GCP Cloud Run, with idempotent, reconciled orders so a retry can't place the same trade twice.
- A React dashboard tracks live positions, P&L and a leaderboard of how each strategy is doing.