Build a COSMIC Cross-Bank Anti-Money Laundering Pipeline with LangChain
Financial institutions require automated threat sharing without compromising customer privacy regulations under Singapore's COSMIC framework. Build a stateful LangGraph workflow using Vercel AI Gateway to aggregate cross-bank suspicious activity reports into standardized, audit-ready COSMIC intelligence assessments.
What you are building
The core problem, expected build, and operating context for this challenge.
Construct a stateful LangGraph workflow routed through Vercel AI Gateway to aggregate cross-bank suspicious activity reports into standardized COSMIC risk evaluations.
How work is evaluated
Evaluates the LangGraph pipeline's ability to ingest transaction logs, redact PII, and generate structured COSMIC risk reports matching target ground truth.
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.
pii_redaction_check
Verifies no raw customer personal data is leaked in the output payload
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
accuracy
F1 score of identified suspicious accounts against ground truth • target: 0.85 • 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
Design a stateful multi-step graph workflow with LangChain and LangGraph
Route multi-model LLM calls through Vercel AI Gateway with fallback policies
Implement data sanitization and privacy-preserving entity resolution for banking records
Generate structured COSMIC-compliant AML threat indicators with high precision
Reference links and supporting material
Synthetic multi-bank ledger datasets containing anomalous payment flows and obfuscated identity markers.
Challenge setup
Evaluates the LangGraph pipeline's ability to ingest transaction logs, redact PII, and generate structured COSMIC risk reports matching target ground truth.
Challenge input
JSON array of raw multi-bank transaction objects with suspect account IDs
LangChain
Provides stateful node execution via LangGraph
Vercel AI Gateway
Offers unified model routing and caching
Evaluated output
JSON object containing risk_score, flagged_entities, privacy_status, and cosmic_report_payload
- Verifies no raw customer personal data is leaked in the output payload
- F1 score of identified suspicious accounts against ground truth • target: 0.85 • range: 0-1
- Benchmark: FinRisk-Bench-2026
- Accuracy target: 0.85
- 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
- LangChain · Required
- Langchain · Optional
- Vercel AI Gateway · Optional
- LangChain · Required
- Langchain · Optional
- Vercel AI Gateway · Optional
Evaluation contract
- pii_redaction_check · Weight 1
- accuracy · Weight 1
Metrics
- accuracy · Target 0.85
Benchmarks
- FinRisk-Bench-2026
- security
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON array of raw multi-bank transaction objects with suspect account IDs
Output: JSON object containing risk_score, flagged_entities, privacy_status, and cosmic_report_payload
Execution harness: Python
Evaluation tools
- LangChain · Agent Orchestration Framework
- Vercel AI Gateway · AI Gateway & Infrastructure
Recipe freeze marker
No recipe freeze time is recorded.