评估驱动开发
Development practice where systematic evals and metrics drive iteration decisions for LLM products.
Dosu describes how it applies evaluation driven development (EDD) to make its LLM-powered engineering agent more reliable, using LangSmith to monitor and search agent activity at scale. The post frames reliability gains as directly reducing maintenance and support burden for the LangChain team.
LangChain has released Promptim, an experimental open-source library that automates prompt optimization loops. Given an initial prompt, a dataset, and custom evaluators, Promptim iteratively produces refined prompts intended to outperform the original, extending LangChain's evaluation-driven development methodology.