Engineering teams lacked automated test suites to measure retrieval hallucination rates across diverse vector embedding models.
Created an open-source synthetic dataset generator and precision benchmark CLI comparing OpenAI, Cohere, and local sentence-transformers.
We evaluate new project briefs and engineering requirements within 12 hours.