Crypto Financial Anomaly Detection
Build a cutting-edge agentic system using LangGraph to analyze complex crypto financial transaction networks. The system will leverage Gemini 2.5 Pro (specifically its 'Deep Think' mode for advanced reasoning) to identify unusual patterns, potential fraud, or hidden relationships within a simulated cryptocurrency transaction dataset. This challenge focuses on designing graph-based agent workflows, implementing hybrid reasoning, and exploring A2A protocol for secure, verifiable data exchange for compliance or audit purposes. Participants will create agents that can traverse transaction graphs, apply heuristic rules and LLM-powered reasoning to detect anomalies, and generate detailed explanations for flagged transactions. The system should demonstrate how graph-based state management can enhance financial oversight in a highly dynamic asset class.
What you are building
The core problem, expected build, and operating context for this challenge.
Implement a graph-based agent workflow using LangGraph and Gemini 2.5 Pro with Deep Think mode for crypto financial anomaly detection and due diligence.
How work is evaluated
The system will be evaluated on its accuracy in identifying pre-defined crypto financial anomalies within a synthetic transaction graph, the quality of its explanations, and the efficiency of its graph-based reasoning workflow. A simulated A2A interaction for reporting anomalies will also be assessed.
Challenge datasets
Public datasets, plus the private datasets you have access to.
What you should walk away with
Develop agents capable of identifying and explaining suspicious crypto transaction patterns.
Design and implement graph-based agent workflows using LangGraph for financial network analysis.
Implement hybrid instant/deep reasoning for efficient anomaly detection.
Leverage Gemini 2.5 Pro's 'Deep Think' mode for complex pattern matching and anomaly reasoning.
Explore A2A protocol for secure, hypothetical interaction with external audit systems.
Reference links and supporting material
Sample data for 3 tasks
Challenge setup
The system will be evaluated on its accuracy in identifying pre-defined crypto financial anomalies within a synthetic transaction graph, the quality of its explanations, and the efficiency of its graph-based reasoning...
Challenge input
{"transactions_csv_url": "URL_to_CSV"}
Agent execution
No agent tool is listed for this challenge.
Evaluated output
{"graph_db_status": "loaded", "node_count": 1000, "edge_count": 5000}
- Verify successful loading of transaction data into a functional graph database accessible by agents.
- Confirm that Gemini 2.5 Pro's Deep Think mode was explicitly invoked for complex anomaly reasoning.
- Ensure the LangGraph workflow executes without errors and transitions between nodes as designed.
- AnomalyDetectionPrecision target: 85
- 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
- Verify successful loading of transaction data into a functional graph database accessible by agents.
- Confirm that Gemini 2.5 Pro's Deep Think mode was explicitly invoked for complex anomaly reasoning.
- Ensure the LangGraph workflow executes without errors and transitions between nodes as designed.
Metrics
- AnomalyDetectionPrecision · Target 85
- ExplanationClarityScore · Target 4
- A2AReportFormatAdherence · Target 90
- ReasoningEfficiency · Target 180
Workflow tasks
- Task 1
Input: {"transactions_csv_url": "URL_to_CSV"}
Output: {"graph_db_status": "loaded", "node_count": 1000, "edge_count": 5000}
Execution harness: Python
- Task 2
Input: {"anomaly_patterns_config": [{"name": "Smurfing", "description": "Many small transactions to a central address"}], "max_anomalies_to_report": 5}
Output: {"flagged_anomalies": [{"type": "Smurfing", "involved_addresses": ["addr1", "addr2"], "explanation": "LLM-generated explanation"}]}
Execution harness: Python
- Task 3
Input: {"anomaly_details": {"type": "Smurfing", "involved_addresses": [...]}, "audit_agent_did": "did:example:auditor"}
Output: {"a2a_report": {"report_id": "UUID", "timestamp": "ISO_DATETIME", "anomaly_type": "Smurfing", "summary": "Brief summary for auditor", "data_hash": "SHA256_HASH"}}
Execution harness: Python
Recipe freeze marker
No recipe freeze time is recorded.