SAR Narrative Generator with Audit Trail
AI-assisted SAR drafting system for AML workflows with multi-stage LLM validation, strict compliance guardrails, role-based review, and sentence-level traceability for regulator-ready auditability.
Project Overview
SAR Narrative Generator with Audit Trail is an AI-assisted system for regulated financial institutions to help analysts produce clear, consistent, regulator-ready SAR narratives from structured AML alert data.
The platform reduces narrative drafting time while enforcing compliance-safe language and preserving defensible outputs.
How It Works
The workflow starts with AML alert JSON input containing customer, account, transaction, and alert information.
A drafting LLM creates a structured narrative, a regulator-focused LLM validates compliance and wording, and an audit LLM maps each sentence to supporting source data.
A human-in-the-loop review lets analysts edit, approve, and finalize the narrative before submission.
Key Features and Stack
The system includes a multi-stage generation pipeline, hard compliance guardrails, role-based workflows, and full audit logging with traceability.
Frontend uses React 18 with React Router, Tailwind CSS, and Axios while the backend uses Node.js, Express, JWT auth, and RBAC.
LLM workflows use Python with LangChain and LangGraph, backed by PostgreSQL and optional Streamlit/FastAPI interfaces.