Supply Chain Compliance Agent with Adaptive Budgets
Addressing concerns about supply chain integrity, this challenge involves building an intelligent multi-agent system using AutoGen and DSPy. Your system will act as a 'Supply Chain Auditor' to investigate potential compliance breaches related to restricted technologies. Leveraging GPT-5 for its advanced reasoning capabilities, a team of specialized agents (e.g., Data Gatherer, Risk Analyst, Report Generator) will collaborate to analyze vast datasets including simulated financial transactions, shipping manifests, public sanctions lists, and news articles. DSPy will be used to programmatically optimize LLM calls for robust information extraction and verification. The system must employ adaptive thinking budgets to manage computational resources and integrate with an MCP for secure access to sensitive enterprise data and reporting suspected violations. This project emphasizes multi-agent orchestration, ethical AI considerations, and data-driven risk assessment.
What you are building
The core problem, expected build, and operating context for this challenge.
Build an AutoGen multi-agent system with DSPy and GPT-5 to audit supply chains for compliance, employing adaptive thinking budgets and MCP for secure data access.
How work is evaluated
Evaluation will assess the multi-agent system's accuracy in identifying compliance breaches, the efficiency of LLM usage, and the clarity and completeness of generated compliance reports. Collaboration between agents and effective use of DSPy and adaptive budgets will be key.
Challenge datasets
Public datasets, plus the private datasets you have access to.
What you should walk away with
Utilize DSPy to programmatically optimize GPT-5 prompts for robust information extraction and reasoning.
Implement adaptive thinking budgets to manage LLM resource consumption during complex investigations.
Design an MCP for secure integration with simulated enterprise data sources and compliance reporting.
Develop a system for identifying and reporting potential supply chain compliance breaches.
Orchestrate a role-based multi-agent team using AutoGen for investigative tasks.
Reference links and supporting material
Sample data for 1 tasks
Challenge setup
Evaluation will assess the multi-agent system's accuracy in identifying compliance breaches, the efficiency of LLM usage, and the clarity and completeness of generated compliance reports. Collaboration between agents...
Challenge input
{'transaction_logs': [{'transaction_id': '...', 'item': '...', 'quantity': ..., 'vendor_id': '...', 'destination_country': '...', 'amount': ...}],...
Agent execution
No agent tool is listed for this challenge.
Evaluated output
{'audit_id': '...', 'status': 'completed', 'breaches_detected': [{'transaction_id': '...', 'item': '...', 'reason': '...', 'evidence_summary': '......
- Verifies that agents retrieve simulated sensitive data only through the MCP with proper authentication.
- Checks if the generated audit report adheres to the specified JSON output format.
- Confirms that DSPy assertions (e.g., output formats, extracted entity types) are met consistently.
- BreachDetectionRecall target: 0.9
- 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
- Verifies that agents retrieve simulated sensitive data only through the MCP with proper authentication.
- Checks if the generated audit report adheres to the specified JSON output format.
- Confirms that DSPy assertions (e.g., output formats, extracted entity types) are met consistently.
Metrics
- BreachDetectionRecall · Target 0.9
- ReportCompletenessScore · Target 0.85
- AdaptiveBudgetEfficiency · Target 0.8
Workflow tasks
- Task 1
Input: {'transaction_logs': [{'transaction_id': '...', 'item': '...', 'quantity': ..., 'vendor_id': '...', 'destination_country': '...', 'amount': ...}], 'vendor_data': [{'vendor_id': '...', 'name': '...', 'country': '...', 'owner_entity': '...'}], 'sanctions_list': [{'entity_name': '...', 'country': '...'}], 'news_articles': [{'title': '...', 'content': '...'}]}
Output: {'audit_id': '...', 'status': 'completed', 'breaches_detected': [{'transaction_id': '...', 'item': '...', 'reason': '...', 'evidence_summary': '...'}], 'risk_score': 0.8, 'recommendations': ['...', ...], 'report_summary': '...', 'llm_tokens_total': 15000}
Execution harness: Python
Recipe freeze marker
No recipe freeze time is recorded.