Multi-Agent Anti-Money Laundering Detection with CrewAI and Kore.ai
Financial institutions in Singapore face complex multi-jurisdictional money laundering threats under the COSMIC information-sharing initiative. Build a CrewAI multi-agent system integrated with Kore.ai workflows to triage shared risk profiles, aggregate cross-bank transaction anomalies, and generate MAS-compliant audit logs with 95% classification precision.
What you are building
Orchestrate role-playing CrewAI agents and Kore.ai workflow actions to parse COSMIC threat intelligence and identify cross-institutional money laundering indicators.
How work is evaluated
Evaluates CrewAI agent group performance on detecting simulated cross-bank transaction anomaly clusters and producing valid MAS risk reports.
Datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
schema_validation
Ensures output JSON matches MAS regulatory reporting structure
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
classification_accuracy
Ratio of correctly identified ML/TF patterns • target: 0.95 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Learning goals
Orchestrate multi-agent Crews with custom roles and tools in CrewAI
Integrate Kore.ai workflow automation nodes for external system actions
Parse and aggregate structured COSMIC ML/TF risk data across banking entities
Generate regulatory-compliant audit trails for MAS reporting
Resources
Synthetic dataset of multi-bank transaction logs and entity risk scores formatted according to MAS COSMIC data exchange specifications.
Challenge setup
Evaluates CrewAI agent group performance on detecting simulated cross-bank transaction anomaly clusters and producing valid MAS risk reports.
Challenge input
JSON containing shared COSMIC transaction graph and entity risk logs
CrewAI
Orchestrates specialized compliance agents
Kore.ai
Automates downstream notification and enterprise system triggers
Evaluated output
JSON containing risk classification, confidence score, and regulatory escalation report
- Ensures output JSON matches MAS regulatory reporting structure
- Ratio of correctly identified ML/TF patterns • target: 0.95 • range: 0-1
- Benchmark: Financial Risk Multi-Agent Benchmark
- Classification Accuracy target: 0.95
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Shows the first task, up to three tools, and up to three evidence items.
View technical recipe
Configured tools
- Kore.ai · Required
- CrewAI · Optional
- Prophet · Optional
Evaluation contract
- schema_validation · Weight 1
- classification_accuracy · Weight 1
Metrics
- classification_accuracy · Target 0.95
Benchmarks
- Financial Risk Multi-Agent Benchmark
- security
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON containing shared COSMIC transaction graph and entity risk logs
Output: JSON containing risk classification, confidence score, and regulatory escalation report
Execution harness: Python
Evaluation tools
- CrewAI · Agent Framework
- Kore.ai · Workflow Automation
Recipe freeze marker
No recipe freeze time is recorded.