Golden Dataset Curation

This skill provides a robust framework for building and refining 'golden datasets'—the ground-truth data used to evaluate AI performance. It implements a sophisticated multi-agent pipeline that automatically fetches content, classifies its type and difficulty, generates test queries, and assigns quality scores across dimensions like accuracy and depth. By integrating with Langfuse for observability and using consensus-based decision making, it ensures that only high-integrity data enters your evaluation pipeline, making it essential for developers building production-grade RAG systems or fine-tuning LLMs.

主要功能

使用场景