Collaborative AML Entity Resolution Swarm with CrewAI and Replit Agent
Financial crime investigators spend thousands of hours piecing together complex, obfuscated networks of shell companies and beneficial owners across global jurisdictions. True autonomy in financial crime compliance requires strict verification safeguards, separating raw reasoning capacity from authority limits. In this challenge, you will construct a multi-agent investigation team using CrewAI and Replit Agent tooling. The swarm will assign distinct investigator roles (e.g., Graph Data Analyst, Sanctions Specialist, Case Compliance Manager) to resolve entity aliases, link suspicious transaction graph nodes, and construct a audit-proof Suspicious Activity Report (SAR).
What you are building
The core problem, expected build, and operating context for this challenge.
Implement a CrewAI multi-agent swarm to perform complex AML entity resolution and suspicious activity reporting.
How work is evaluated
Evaluates entity resolution precision, recall of linked shell companies, and compliance gate enforcement.
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.
authority_gate_test
Ensures no auto-filing occurs without human authorization flag
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
f1_score_entity_matching
F1 score of entity link detection in multi-hop networks • target: 0.92 • 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
Define specialized CrewAI agents with clear permissions and tool sets
Build custom tool bindings to perform graph query lookups on transaction nodes
Enforce human-in-the-loop approval gates before finalizing formal SAR filings
Measure entity resolution accuracy across noisy corporate registry data
Reference links and supporting material
Synthetic multi-jurisdictional financial node dataset containing obfuscated entity names, corporate registries, and circular transaction rings.
Challenge setup
Evaluates entity resolution precision, recall of linked shell companies, and compliance gate enforcement.
Challenge input
JSON representing graph of account transfers and company registrations
CrewAI
Role-based collaborative agent orchestration platform.
Replit Agent
Rapid prototyping environment for agentic applications.
Evaluated output
JSON with resolved_entities, linked_cluster_id, and compliance_verdict
- Ensures no auto-filing occurs without human authorization flag
- F1 score of entity link detection in multi-hop networks • target: 0.92 • range: 0-1
- Benchmark: FinCrime-Bench-Graph
- F1 Score Entity Matching target: 0.92
- 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
- CrewAI · Required
- crewAI · Optional
Evaluation contract
- authority_gate_test · Weight 1
- f1_score_entity_matching · Weight 1
Metrics
- f1_score_entity_matching · Target 0.92
Benchmarks
- FinCrime-Bench-Graph
- security
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON representing graph of account transfers and company registrations
Output: JSON with resolved_entities, linked_cluster_id, and compliance_verdict
Execution harness: Python
Evaluation tools
- CrewAI · Agent Framework
- Replit Agent · AI Engineering Tooling
Recipe freeze marker
No recipe freeze time is recorded.