Realtime fraud alert explainer with evidence trails

Project to turn fraud model alerts into human-readable stories. Design: - Ingest: Streaming alerts from rules and ML models. - Retrieve: Fetch relevant customer history and similar past cases. - Explain: LLM converts features into narratives with confidence and caveats. - Actions: Suggest next steps and required approvals. - Audit: Immutable logs of data used and explanations shown. - Safety: PII masking; strict access controls. Improves investigator efficiency and reduces false positives through context.