AI Governance Engineer

The AI Governance Engineer ensures the safe, effective, and ethical deployment of AI/ML across the FHSC system. This role operationalizes the governance framework for the full AI lifecycle, conducts technical due diligence of internal and external AI systems, and establishes robust post-implementation monitoring and auditing to safeguard patient care, regulatory compliance, and institutional integrity. The engineer partners closely with clinical leaders, data scientists, IT, Legal, Compliance, Safety/Risk, and Procurement to embed responsible AI practices in both vendor and in-house solutions.

Essential Functions:

• Evaluation and validation of external/vendor AI tools.
• Lead pre-implementation risk assessments for commercial AI/ML tools used in clinical, research, and administrative contexts.
• Assess vendor documentation for alignment with FHSC governance standards, ethical principles, and regulatory requirements (e.g., FDA SaMD guidance and GMLP, HIPAA, ONC regulations, NIST AI RMF, emerging AI/LLM standards).
• Validate vendor claims and performance: generalizability to FHSC populations, calibration, bias/fairness across subgroups, robustness, transparency, and explain ability within FHSC data and workflows.
• Evaluate LLM-specific risks (e.g., hallucination rates, prompt injection/jailbreaks, data leakage/PII exposure, harmful/biased outputs) and verify mitigations.
• Internal AI architecture and framework design.
• Assess existing AI/ML technical architectures, data pipelines, and MLOps workflows used by clinical and research teams; identify gaps and remediation plans.
• Co-develop and maintain SOPs covering the full AI lifecycle: data sourcing and quality, labeling, privacy and de-identification, feature engineering, model development, validation, documentation, versioning, deployment, change management, and retirement.
• Define and promote Responsible AI practices: bias detection/mitigation, explainability, human-in-the-loop oversight, safety testing, and reproducibility. Advise teams on privacy-by-design, security-by-design, and safety-by-design, including PHI minimization, access controls, and differential privacy where appropriate.
• Ongoing monitoring and algorithm vigilance Design, implement, and manage a continuous monitoring framework (algorithm vigilance) for all deployed AI/ML systems.
• Define and track KPIs and risk indicators: accuracy, calibration, sensitivity/specificity, AUROC/PR, subgroup performance/fairness, model/data drift, latency, uptime, hallucination/toxicity rates (for LLMs), override rates, and clinical/operational impact.
• Build automated dashboards and alerting for real-time issue detection; implement thresholds, canary releases, rollback/kill-switch and fallback procedures.
• Coordinate with Safety and Risk on incident triage, root-cause analysis, corrective action plans, and post-incident reviews; run periodic tabletop exercises.
• Policy, governance, and reporting Contribute to FHSC's AI Governance Policy, Risk Management Framework, and associated standards, aligning with NIST AI RMF, ISO/IEC 23894, FDA/IMDRF SaMD, HIPAA, and emerging regulations.
• Prepare and present risk assessments, validation reports, and operational metrics to the AI Governance Committee, executive leadership, and clinical/operational stakeholders.

Qualifications

• Bachelor's Degree Computer Science, Data Science, Engineering, or other related field.
• 3-6 years of experience in AI/ML model validation, AI observability/evaluations, technical risk management, or governance in a regulated environment.

Technical Knowledge, Skills, and Abilities:

• Strong understanding of ML fundamentals and MLOps/model lifecycle management, statistical evaluation, data governance, and LLM evaluation and risk mitigation (e.g., benchmarking, robustness, bias/toxicity, hallucination).
• Familiarity with regulatory frameworks relevant to health AI: FDA SaMD and GMLP, HIPAA, ONC/HTI-1; and industry frameworks such as NIST AI RMF (and awareness of emerging rules).
• Demonstrated ability to translate complex technical findings about performance, bias, and architecture into clear, actionable insights for non-technical stakeholders. Proficiency with Python and SQL, and with tools for experimentation, versioning, and monitoring (e.g., MLflow/Weights & Biases, Evidently/WhyLabs/Fiddler/Arize, Great Expectations, Git).