Autonomous Crypto Compliance Agent
This challenge requires building an advanced autonomous agent focused on financial compliance within the cryptocurrency domain or complex supply chain networks. Utilizing OpenAI's Agent SDK, developers will create a system capable of real-time monitoring of transactions and identifying suspicious patterns indicative of illicit activities or regulatory breaches. The agent will leverage sophisticated tool use and function calling to interact with external data sources and analytical frameworks. The core of the system involves GPT-5-2 for advanced reasoning and orchestrating analytical tasks. It will integrate Darts, a time-series forecasting library, to detect anomalies in transaction volumes or patterns over time. Long-term memory and regulatory context will be managed by a vector database (e.g., Pinecone) storing an extensive knowledge base of financial regulations and compliance policies. The agent will also interact with a simulated Enterprise Transaction API to fetch real-time data. The goal is to develop an intelligent agent that not only identifies potential compliance issues but also provides detailed reports, evidence, and recommendations for further investigation, showcasing a modern, proactive approach to financial crime detection.
What you are building
The core problem, expected build, and operating context for this challenge.
Develop an autonomous financial compliance agent using OpenAI Agents SDK, GPT-5-2, Darts for time-series analysis, and a vector database for regulatory memory.
How work is evaluated
The evaluation will assess the agent's accuracy in detecting predefined suspicious financial activities, the relevance and comprehensiveness of its compliance reports, and the efficiency of its tool usage.
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.
AccuracyOfSuspiciousDetection
Agent must correctly identify at least 90% of predefined suspicious transaction patterns.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
ReportCompleteness
Generated reports must include a summary, flagged transactions, and recommendations for at least 95% of cases.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Anomaly Detection Precision
Precision score for identifying true anomalies among all flagged transactions. • target: 0.85 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Tool Call Efficiency
Average number of tool calls per detected suspicious event, aiming for minimal yet effective calls. • target: 2.5 • range: 1-5
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Master the OpenAI Agents SDK for constructing autonomous, tool-enhanced agents capable of complex decision-making and interaction.
Implement advanced function calling within the agent's workflow to integrate with external systems, including Darts for time-series analysis.
Utilize Darts for building sophisticated time-series models to detect anomalies in cryptocurrency transaction data, identifying potential illicit activities.
Design and manage a long-term memory system using Pinecone vector database to store and retrieve relevant financial regulations and compliance policies for dynamic context.
Orchestrate GPT-5-2's reasoning capabilities to interpret complex regulatory text, evaluate transaction data, and generate compliance reports with actionable insights.
Build a robust data ingestion pipeline to feed real-time transaction data from a simulated Enterprise Transaction API into the agent for continuous monitoring.
Develop strategies for evaluating agent performance in detecting fraud patterns and minimizing false positives in a dynamic financial environment.
Reference links and supporting material
Sample data for 2 tasks
Challenge setup
The evaluation will assess the agent's accuracy in detecting predefined suspicious financial activities, the relevance and comprehensiveness of its compliance reports, and the efficiency of its tool usage.
Challenge input
{ "transaction_stream": [ { "id": "T101", "sender": "A", "receiver": "B", "amount": 1000, "timestamp": "..." }, ... ], "regulatory_context": "[rele...
Agent execution
No agent tool is listed for this challenge.
Evaluated output
{ "is_suspicious": true, "reason": "[explanation]", "flagged_transactions": ["T105", "T106"], "compliance_breach": "[rule name]", "recommendations"...
- Agent must correctly identify at least 90% of predefined suspicious transaction patterns.
- Generated reports must include a summary, flagged transactions, and recommendations for at least 95% of cases.
- Precision score for identifying true anomalies among all flagged transactions. • target: 0.85 • range: 0-1
- Anomaly Detection Precision target: 0.85
- 2 public reference cases
- Python execution harness
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
No tool records are attached.
Evaluation contract
- AccuracyOfSuspiciousDetection · Weight 1
- ReportCompleteness · Weight 1
- Anomaly Detection Precision · Weight 1
- Tool Call Efficiency · Weight 1
Metrics
- Anomaly Detection Precision · Target 0.85
- Tool Call Efficiency · Target 2.5
Reference cases
- 2 public reference cases
Workflow tasks
- Task 1
Input: { "transaction_stream": [ { "id": "T101", "sender": "A", "receiver": "B", "amount": 1000, "timestamp": "..." }, ... ], "regulatory_context": "[relevant Pinecone query result]" }
Output: { "is_suspicious": true, "reason": "[explanation]", "flagged_transactions": ["T105", "T106"], "compliance_breach": "[rule name]", "recommendations": ["action"] }
Execution harness: Python
- Task 2
Input: { "suspicious_event_details": { "type": "structuring", "flagged_transactions": [...] }, "relevant_regulations": "[Pinecone query result]" }
Output: { "report_summary": "[text]", "evidence_links": ["url"], "risk_assessment": "High" }
Execution harness: Python
Recipe freeze marker
No recipe freeze time is recorded.