Problem
AML investigators face noisy alerts and need case-ready evidence, risk reasoning, and reviewable outputs rather than opaque automation.
FeaturedAI / LLM SystemsMachine LearningFinancial Services
AML Agentic Intelligence Engine
Multi-agent AML investigation system for alert triage, evidence assembly, risk reasoning, and SAR/STR drafting.
Approach
Designed a governed multi-agent workflow with explicit state, retrieval, guardrails, audit logs, and evaluator checks around investigator review.
Outcome
Framed an applied AI system around measurable alert quality, high-risk detection, invalid alert reduction, and human oversight.
Technical Stack
LangGraphRAGGuardrailsAgent StateAudit LoggingLLM-as-JudgePython
Metrics & Signals
15% high-risk detection improvement20% invalid alert reductionHuman-in-the-loop review