Multi-Agent Sanctions Compliance Tracker using CrewAI and Fireworks AI
Rapid shifts in international sanctions lists require immediate reconfiguration of transaction screening parameters for global payment networks. Build a multi-agent compliance squad with CrewAI and Fireworks AI inference to ingest regulatory updates, re-index sanction tables, and evaluate suspended transactions. Achieve 100% detection of sanctioned entities with zero latency degradation on cleared routes.
What you are building
The core problem, expected build, and operating context for this challenge.
Orchestrate a crew of specialized compliance agents using CrewAI and Fireworks AI to continuously ingest regulatory updates and screen pending payments.
How work is evaluated
Tests crew efficiency in updating sanction policies and correctly updating transaction release/block decisions.
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.
policy_update_test
Ensures status switches to RELEASED upon lifting of sanctions
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
sanction_accuracy
Accuracy of transaction status determinations after regulatory changes • target: 1 • 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 roles: Regulatory Crawler, Rule Reindexer, and Payment Evaluator
Configure Fireworks AI high-speed inference endpoints for agent task execution
Implement multi-agent sequential and hierarchical task delegation patterns
Demonstrate live dynamic update of payment routing decisions following sanction status changes
Reference links and supporting material
Dataset of 20 regulatory bulletin announcements and 300 cross-border pending transactions in sanction-transition zones.
Challenge setup
Tests crew efficiency in updating sanction policies and correctly updating transaction release/block decisions.
Challenge input
JSON containing news text regarding sanction modification and held transaction queue
CrewAI
Agent collaboration framework
Fireworks AI
Fast model inference runner
Evaluated output
JSON array of transaction IDs with updated statuses (RELEASED or BLOCKED)
- Ensures status switches to RELEASED upon lifting of sanctions
- Accuracy of transaction status determinations after regulatory changes • target: 1 • range: 0-1
- Benchmark: Sanctions-Eval
- Sanction Accuracy target: 1
- 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
- Fireworks AI · Optional
- Fireworks AI · Optional
Evaluation contract
- policy_update_test · Weight 1
- sanction_accuracy · Weight 1
Metrics
- sanction_accuracy · Target 1
Benchmarks
- Sanctions-Eval
- security
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON containing news text regarding sanction modification and held transaction queue
Output: JSON array of transaction IDs with updated statuses (RELEASED or BLOCKED)
Execution harness: Python
Evaluation tools
- CrewAI · Multi-Agent Framework
- Fireworks AI · Inference Engine
Recipe freeze marker
No recipe freeze time is recorded.