Generative AI Medical Device Regulatory Compliance Checker with AutoGen
FDA regulatory oversight for generative AI medical devices demands strict compliance auditing of model risk controls and lifecycle logs. You will build an AutoGen multi-agent system featuring regulatory auditor and medical device engineer agents to verify AI clearance documentation with 92% regulatory alignment.
What you are building
Design an AutoGen multi-agent debate and validation loop to audit generative AI medical device submissions against FDA regulatory frameworks.
How work is evaluated
Evaluates regulatory compliance deficiency detection accuracy in SaMD documentation.
Datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
gap_identification_check
Ensure critical regulatory gaps are detected
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
compliance_precision
Precision of identified FDA regulatory non-compliance gaps • target: 0.92 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Learning goals
Implement multi-agent conversation systems using Microsoft AutoGen
Create specialized conversable agents for FDA AI/ML medical device guidance
Automate verification of AI drift controls, risk mitigation plans, and software logs
Resources
Dataset of 100 generative AI medical device software documentation files with annotated regulatory non-compliance flags.
Challenge setup
Evaluates regulatory compliance deficiency detection accuracy in SaMD documentation.
Challenge input
JSON object with document_text string
AutoGen
Mandatory Microsoft AutoGen framework.
Evaluated output
JSON object with is_compliant boolean and identified_gaps list
- Ensure critical regulatory gaps are detected
- Precision of identified FDA regulatory non-compliance gaps • target: 0.92 • range: 0-1
- Benchmark: RegAI-Bench
- Compliance Precision target: 0.92
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Shows the first task, up to three tools, and up to three evidence items.
View technical recipe
Configured tools
- AutoGen · Required
- Aim · Optional
- Kombai · Optional
Evaluation contract
- gap_identification_check · Weight 1
- compliance_precision · Weight 1
Metrics
- compliance_precision · Target 0.92
Benchmarks
- RegAI-Bench
- general
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON object with document_text string
Output: JSON object with is_compliant boolean and identified_gaps list
Execution harness: Python
Evaluation tools
- AutoGen · Multi-Agent Framework
Recipe freeze marker
No recipe freeze time is recorded.