Forensic ML debugging console — investigates why a model got a prediction wrong using SHAP attribution, nearest-neighbour forensics, confidence decomposition, and an AI-written failure report across credit risk, medical, and equipment datasets.

Production-grade anomaly detection across 5 correlated IoT signals using triple-layer detection (Z-score / IQR / CUSUM) and Granger causality for root cause ranking. Features an evidence timeline and AI-written incident report.

Bayesian structural time series tool that answers "did this business decision actually work?" — models a statistical counterfactual and measures the true causal lift of interventions with 95% credible intervals across 4 real business scenarios.

Natural language interface over a multi-table relational database using live schema injection and LLM inference. Validates SQL via DuckDB EXPLAIN before execution and returns results with an AI business explanation.

Auto-charting pipeline that profiles any CSV, infers column types, routes each variable to the right Plotly chart, and uses a local LLM to generate plain-English business narratives per chart — zero cloud API required.