Summary
All About AI documents the build and first 24-hour performance of a fully automated trading system for the Hyperliquid crypto platform, constructed using GPT-5.6 (specifically the “Soul” model) via Claude Code. The system scans markets every five minutes, scores trade opportunities across five criteria—liquidity (up to +20 pts), directional momentum (+15), RSI balance (+15), trend width (+20), and bounce strength (+18)—and opens positions automatically when a threshold of 70 points is exceeded. The creator demonstrates a live SpaceX short trade that scored 96 points, ran for 43 minutes, and netted $24 profit after fees, using 10x leverage on a small training account.
The second half of the video shows GPT-5.6 Soul at “max” reasoning level analyzing 24 hours of trade logs to suggest improvements. Rather than overhauling the strategy, the model recommended reliability upgrades: retry logic, emergency exit handling, and improved data labeling with economic correlation clustering to limit simultaneous correlated positions. After implementing the changes, 44 automated tests passed successfully.
The creator frames the system as a living learning loop—collect data, feed it back to GPT-5.6, refine the strategy—emphasizing that the AI builds the code while the human defines the scoring logic and improvement goals. The setup is positioned as a starting template for anyone wanting to experiment with AI-assisted algorithmic trading.
📺 Source: All About AI · Published July 15, 2026
🏷️ Format: Hands On Build
