Composo

Composo builds evaluation and quality systems for production AI. It analyzes application traces to identify failure patterns, organizing findings by their type, severity and frequency.

Domain experts review the findings and correct the system’s judgments. Those examples help calibrate evaluation to the standards of the application, including cases where a generic correctness rubric misses an important error.

The resulting criteria and confirmed failure patterns can be used in runtime guardrails. Composo offers self-hosted deployment and gives teams ownership of their evaluation criteria, correction data and rules.