DataDriftX

DataDriftX is a web app (with optional Slack alerts) that monitors deep-learning models in production for data drift, label drift, and silent performance decay. You connect it to your prediction logs and a small trickle of ground-truth labels (when available). It automatically builds baseline feature distributions, embedding-space monitors, and slice-based dashboards (by region, device, cohort, etc.) to pinpoint where the model is failing. When labels are sparse, it uses proxy signals (confidence shifts, embedding centroid movement, and disagreement between shadow models) to raise “investigate” alerts rather than pretending it knows accuracy. The product is intentionally narrow: monitoring and diagnosis, not full MLOps. It exports concise incident reports for engineers and non-technical stakeholders, and suggests the minimum data to collect next to confirm or refute suspected drift. This is an AI app combined with traditional analytics software.