Build a COSMIC Anti-Money Laundering Sharing Agent with Google ADK
Financial crime teams struggle to cross-analyze collaborative intelligence without exposing raw PII. Build a Google ADK agent integrated with Helicone observability that parses anonymized COSMIC sharing logs to flag cross-bank structuring patterns, achieving a 95% detection accuracy on synthetic financial intelligence datasets.
What you are building
Develop an anti-money laundering analytics agent using Google ADK and Helicone to analyze collaborative banking signals without privacy leakage.
How work is evaluated
Evaluates the agent's accuracy in identifying structured AML rings and compliance with audit logging.
Datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
schema_validation
Ensures output conforms to the structured JSON risk schema.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
detection_precision
Precision score of correctly identified AML network rings. • 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
Master privacy-preserving data extraction using Google ADK tools
Implement multi-bank transaction graph reconstruction for AML monitoring
Integrate Helicone observability for audit tracing and latency optimization
Build regulator-friendly explainable anomaly detection reports
Resources
Synthetic multi-bank inter-institution financial log dataset mirroring COSMIC data structures.
Challenge setup
Evaluates the agent's accuracy in identifying structured AML rings and compliance with audit logging.
Challenge input
JSON formatted inter-bank transaction logs
Google ADK
Agent orchestration with native Gemini capabilities.
Helicone
Tracing prompt outputs and latency for regulatory compliance.
Evaluated output
JSON risk assessment report with flagged account IDs and confidence scores
- Ensures output conforms to the structured JSON risk schema.
- Precision score of correctly identified AML network rings. • target: 0.95 • range: 0-1
- Benchmark: FinCrimeAgentBench
- Detection Precision 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
- Google ADK · Required
- Google · Optional
- Helicone · Optional
Evaluation contract
- schema_validation · Weight 1
- detection_precision · Weight 1
Metrics
- detection_precision · Target 0.95
Benchmarks
- FinCrimeAgentBench
- security
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON formatted inter-bank transaction logs
Output: JSON risk assessment report with flagged account IDs and confidence scores
Execution harness: Python
Evaluation tools
- Google ADK · Agent Framework
- Helicone · Observability
Recipe freeze marker
No recipe freeze time is recorded.