Building an AI Crypto Trading Workflow with Multi-Timeframe Signals

Summary

The article walks through an automated crypto perpetual-futures system that gathers account, market, and position data, sends a structured context to a large language model, and routes its output to an execution module. Its market inputs combine short-interval and four-hour candles with moving averages, MACD, RSI, ATR, volume, funding, and recent price series. The workflow also tracks open positions and exit plans, then uses machine-readable signals to place or close trades.

The author describes prompt constraints for position limits, avoiding additions to existing positions, maximum trade risk, minimum reward-to-risk, and required stops, targets, and invalidation conditions. The article offers code fragments and a conceptual example of using the longer timeframe for direction and the shorter timeframe for timing. It reports a competition among AI models as motivation, but does not present controlled performance analysis of the replicated system. Limitations include event blindness, fixed polling intervals, leverage and slippage exposure, and variable model outputs; historical validation is identified as a needed improvement.

Key ideas

• The system separates market and account data collection, model decisions, and order execution.
• Short- and longer-timeframe indicators serve different roles in the decision process.
• Structured model output and position limits make automated signal handling easier to constrain.
• Every position is intended to include a stop, target, and invalidation condition.
• The described workflow lacks controlled performance evidence and remains exposed to model and market risks.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.