How to build prediction trading market bots with Claude

AI-Powered Prediction-Market Trading Bot

The video describes a proposed prediction-market trading bot that can be built using a “Claude” or cloud skill. The presenter claims the system has a 68.4% win rate and could make $1,500 per day. These promotional claims are not independently verified in the subtitles.

Five-Step Workflow

1. Scan markets: Filter more than 300 active markets by liquidity, volume, and time to resolution. Flag unusual price movements or wide spreads.
2. Research in parallel: Agents gather information from Twitter, Reddit, and RSS feeds, analyze sentiment, and compare the prevailing narrative with market odds.
3. Estimate probabilities: A prediction agent combines XGBoost with a large language model to estimate a market’s “true” probability. It trades only when confidence exceeds a threshold.
4. Manage risk and trade: A risk agent sizes positions based on the estimated edge and bankroll, blocks trades it judges too risky, and monitors accepted on-chain trades until settlement.
5. Review losses: Five agents conduct postmortems after a loss, record what went wrong, and update the system to reduce repeated mistakes.

The presenter also refers to a 33-page guide or cheat sheet on building these skills and offers a customized AI automation service.

Main source: Ray Fu, the video’s presenter.