Description

This project utilizes DSPy to decouple task contracts, execution, and evaluation from prompt search, transitioning from manual prompt engineering to a systematic scored optimization loop. The architecture allows for iterative improvements in legal analysis quality, specifically addressing failures where AI answers look authoritative but lack valid underlying evidence. The system features automated evidence checking, confidence scoring, and source URL verification.

Descriptions, tags, and model credits may be AI-generated or inferred from public sources and can be incomplete or wrong. Learn more.