RecAudit

RecAudit is a web app that continuously tests and explains your recommendation system’s real business impact. Instead of trusting offline metrics, it plugs into your event stream (views, clicks, add-to-cart, purchases) and runs lightweight, always-on experiments: holdouts, interleaving, and segmented A/B tests. It flags when a model is “winning” only on vanity metrics, when it harms long-tail discovery, or when it over-personalizes into filter bubbles. The AI layer summarizes results in plain language, suggests next experiments, and highlights likely causes (data drift, cold-start issues, popularity bias, seasonality). You get dashboards for lift, revenue per session, diversity, fairness proxies, and latency/uptime. This is not a magic model-builder; it’s a measurement and governance layer that helps teams stop shipping regressions and defend decisions to stakeholders.