Multi-Agent AML Case Triage and Investigation System with AutoGen
Anti-Money Laundering (AML) investigators face overwhelming backlogs of false-positive transaction alerts from legacy monitoring systems. Build a multi-agent conversational investigation team using AutoGen to synthesize transaction histories, cross-reference KYC records, and generate audit-ready SAR reports. Reduce investigator manual triage time by 60% while maintaining 100% regulatory audit compliance.
What you are building
The core problem, expected build, and operating context for this challenge.
Build a multi-agent AML case triage system using AutoGen to coordinate specialized investigator agents and automate Suspicious Activity Report drafting.
How work is evaluated
Evaluates multi-agent conversation flow, SAR narrative generation quality, and false-positive dismissal accuracy.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
structuring_detection_test
Ensures sub-$10k repeat deposits are flagged as SAR_REQUIRED
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
investigation_accuracy
Accuracy of case dispositions across ground truth test set • target: 0.98 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Construct multi-agent AutoGen conversations featuring Data Retriever, KYC Reviewer, and Lead Investigator
Synthesize structured bank transaction logs with historical account behavior
Automate generation of compliant Suspicious Activity Reports (SAR) narratives
Incorporate Human-in-the-Loop review gates for final SAR sign-off
Reference links and supporting material
Dataset of 200 transaction alert cases with associated KYC history, including false positives and structuring patterns.
Challenge setup
Evaluates multi-agent conversation flow, SAR narrative generation quality, and false-positive dismissal accuracy.
Challenge input
JSON transaction alert record including account history and counterparty details
AutoGen
Microsoft conversational multi-agent framework
Evaluated output
JSON containing case_disposition (SAR_REQUIRED or FALSE_POSITIVE), narrative_text, and audit_trail
- Ensures sub-$10k repeat deposits are flagged as SAR_REQUIRED
- Accuracy of case dispositions across ground truth test set • target: 0.98 • range: 0-1
- Benchmark: AMLCase-Eval
- Investigation Accuracy target: 0.98
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
- AutoGen · Required
- Microsoft AutoGen · Optional
Evaluation contract
- structuring_detection_test · Weight 1
- investigation_accuracy · Weight 1
Metrics
- investigation_accuracy · Target 0.98
Benchmarks
- AMLCase-Eval
- security
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON transaction alert record including account history and counterparty details
Output: JSON containing case_disposition (SAR_REQUIRED or FALSE_POSITIVE), narrative_text, and audit_trail
Execution harness: Python
Evaluation tools
- AutoGen · Agent Framework
Recipe freeze marker
No recipe freeze time is recorded.