TraceLens

TraceLens is a web app (with optional desktop agent) that turns “black box” ML predictions into auditable interpretability reports your team can actually trust. It ingests a model endpoint (or batch predictions), the input schema, and a sample dataset, then generates consistent explanations using multiple methods (SHAP, Integrated Gradients, counterfactuals) and flags when methods disagree. It focuses on the unsexy but real need: producing repeatable artifacts for reviews, incident postmortems, and compliance. Each report includes data drift context, feature attribution stability over time, and a “what would change the outcome” section with guardrails to avoid nonsensical counterfactuals. It also logs explanation versions so teams can compare before/after a retrain. This is an AI app + traditional app: AI helps summarize findings and generate human-readable narratives, but the core value is rigorous, reproducible interpretability pipelines.