RecAudit

RecAudit is a web app (with optional API) that continuously tests and monitors recommendation systems for quality, bias, and revenue-impacting failure modes. Teams connect their event logs (clicks, views, purchases) and model outputs, then RecAudit runs automated offline evaluations, cohort-based fairness checks, and “what changed?” drift analysis after each model release. It produces plain-English reports for product leaders and detailed diagnostics for ML engineers: coverage gaps, over-personalization, popularity traps, cold-start pain, and segment-level regressions. It also generates synthetic test users and counterfactual scenarios to stress-test edge cases before deployment. This is an AI + traditional app: AI helps generate tests, summarize findings, and suggest fixes, while the core value is rigorous evaluation and monitoring. The product is realistic: it won’t magically improve your model, but it will stop you from shipping silent regressions that cost money and trust.