Skip to main content
Loading navigation
Back to challenges

FDA AI Medical Device Safety Compliance Auditor using OpenAI Agents SDK and Fireworks AI

FDA regulation of generative AI medical devices mandates real-time safety validation and hallucination checks on generated diagnostic reports. Build an audit agent using OpenAI Agents SDK and Fireworks AI to continuously monitor generative outputs and flag non-compliant medical device responses with 95% recall.

Data ScienceHosted by Vera
Sign in free to start

What you are building

Build an automated FDA safety compliance audit agent with OpenAI Agents SDK and Fireworks AI high-speed inference.

How work is evaluated

Evaluation

Evaluates recall in detecting non-compliant generative AI diagnostic outputs.

Datasets

Public datasets, plus the private datasets you have access to.

How submissions are scored

Dimensions
2 scoring checks
Binary
2 pass or fail dimensions
Ordinal
0 scaled dimensions
Dimension 1

audit_output_validity

Ensures audit reports flag mandatory violations accurately.

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Dimension 2

violation_recall

Recall metric for detecting unsafe medical advice in test corpus. • target: 0.95 • range: 0-1

Binary check

This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.

Learning goals

  • Utilize OpenAI Agents SDK for multi-turn autonomous compliance auditing.

  • Integrate Fireworks AI for ultra-low latency safety check execution.

  • Implement real-time regulatory compliance rule validation for generative AI devices.

  • Achieve high recall on detecting unsafe or hallucinated clinical recommendations.

Resources

Dataset notes

Synthetic dataset of medical device generative outputs annotated with FDA GMLP safety non-compliance flags.

Challenge setup

Evaluates recall in detecting non-compliant generative AI diagnostic outputs.

Challenge input

JSON containing candidate AI radiology summary.

OpenAI Agents SDK

Official framework for multi-turn function calling agents.

Fireworks AI

High-throughput, low-latency LLM inference provider.

Evaluated output

JSON with compliance_pass (boolean) and list of flagged_violations.

Evaluation checks
  • Ensures audit reports flag mandatory violations accurately.
  • Recall metric for detecting unsafe medical advice in test corpus. • target: 0.95 • range: 0-1
Configured evaluation evidence
  • Benchmark: Medical AI Regulatory Safety Benchmark
  • Violation Recall target: 0.95
  • 1 public reference case
Configured runtime
  • 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

Action Space
  • OpenAI · Required
  • Aim · Optional
Policy Serving
  • Fireworks AI · Optional
Training Infra
  • Fireworks AI · Optional

Evaluation contract

  • audit_output_validity · Weight 1
  • violation_recall · Weight 1

Metrics

  • violation_recall · Target 0.95

Benchmarks

  • Medical AI Regulatory Safety Benchmark
  • security

Reference cases

  • 1 public reference case

Workflow tasks

  1. Task 1

    Input: JSON containing candidate AI radiology summary.

    Output: JSON with compliance_pass (boolean) and list of flagged_violations.

    Execution harness: Python

Evaluation tools

  • OpenAI Agents SDK · Agent Framework
  • Fireworks AI · Inference Runtime

Recipe freeze marker

No recipe freeze time is recorded.

Frequently Asked Questions about FDA AI Medical Device Safety Compliance Auditor using OpenAI Agents SDK and Fireworks AI