Show HN: I made an agent can self-improving while working

Category: ai-ml

Tags: ai-agents, meta-learning, benchmarking, prompt-optimization, typescript

Score: 7.3/10 (Innovation: 7, Technical: 8, Documentation: 7, Utility: 7)

Gear is an open-source optimization framework that uses meta-learning to iteratively improve AI agent performance by evolving prompts, tools, workflows, and model weights based on task evaluation feedback. It demonstrates significant benchmark gains (e.g., 96% relative improvement on AutomationBench) and offers a structured approach to bridge the gap between general models and specific real-world tasks.

Target audience: AI engineers, ML researchers, and developers building AI agents

Repository: https://yehuitang0316.github.io/gear-site/ · TypeScript · MIT · 164 stars

View on Hacker News