pi-autoresearch

pi-autoresearch is an extension for the pi AI coding agent that enables autonomous optimization loops for various performance metrics. It allows the agent to iteratively test ideas, benchmark results, and maintain improvements while automatically reverting regressions. The system provides a live dashboard and confidence scoring to help developers distinguish real performance gains from benchmark noise.

pi-autoresearch is an extension for the pi AI coding agent that implements an autonomous optimization loop, enabling the agent to iteratively test, benchmark, and refine code based on specific performance metrics. By decoupling domain-agnostic infrastructure from domain-specific skills, the project allows developers to automate complex tasks like bundle size reduction or test speed optimization while maintaining session state across restarts. It addresses the challenge of noisy benchmark data by incorporating a confidence scoring system based on Median Absolute Deviation, ensuring that improvements are statistically significant before they are finalized.