TraceLens

TraceLens is a web app (with optional desktop agent) that turns ML/LLM predictions into audit-ready interpretability reports. It ingests model inputs/outputs, feature metadata, and training snapshots, then generates consistent explanations using multiple methods (e.g., SHAP for tabular, Integrated Gradients for deep nets, attention/attribution plus counterfactuals for LLM prompts). The product focuses on repeatability: every explanation is versioned, linked to the exact model artifact, dataset hash, and environment, and can be re-run to reproduce results. It also flags “explanation instability” by measuring how sensitive explanations are to small input perturbations and model updates—because many interpretability tools look convincing but aren’t reliable. Reports export to PDF/JSON for compliance teams and can be embedded into internal model dashboards. This is an AI app + traditional app combination.