AI AI Governance PortfolioTechnical Program Manager · Model Risk · Responsible AI
Website GitHub
AI Governance · Technical Program Manager

Turning AI regulation into running infrastructure.

I build the controls, validation harnesses, and audit pipelines that let organizations deploy AI with confidence — translating NIST AI RMF, the EU AI Act, ISO/IEC 42001 and SR 11-7 into systems engineers can ship and examiners can verify.

View Projects
Selected Work

Projects

Each project targets a different point in the AI lifecycle — from the data entering a model to the business impact of its decisions.

1st line · Build & operateData quality gates and controls embedded in pipelines.
2nd line · Independent oversightModel validation, bias auditing and risk classification.
3rd line · AssuranceEvery project emits audit-ready evidence for internal audit.
LLM Validation · Bias Auditing01

LLM Claims Triage Systems Validation

Second-line model risk governance & bias auditing harness

An independent MRM validation harness for auditing NLP- and LLM-driven claims triage engines. It evaluates generative outputs for non-deterministic drift, demographic disparity, and decision alignment — bringing automated fairness auditing and explainability controls to high-consequence workflows.

  • Detects output drift in non-deterministic LLM decisions
  • Quantifies demographic disparity with Fairlearn metrics
  • Maps findings to NIST AI RMF, NAIC and SR 11-7

Tech stack

PythonFairlearnPandasNumPyTokenization / embeddingsPrompt & output validation

Standards

NIST AI RMF 1.0NAIC Model BulletinSR 11-7 / OCC
Enterprise Risk · Compliance02

ERM-X

Enterprise AI risk management & compliance orchestration framework

An enterprise-grade AI risk governance framework that operationalizes Tier-1 model risk management across production environments. It automates end-to-end risk classification, audit-ready inference logging, and continuous drift monitoring — bridging engineering pipelines with regulatory baselines at scale.

  • Automated risk-tier classification for production models
  • Structured JSON inference logs and Model Card metadata
  • Distributed drift monitoring on Spark / Databricks

Tech stack

PythonSQLPySparkDatabricksStreamlitJSON logging schemasModel Cards

Standards

NIST AI RMF 1.0EU AI ActISO/IEC 42001SR 11-7 / OCC
Data Governance · Lineage03

Socrata–PLUTO QA Audit Gate

Automated data quality & lineage validation for public-sector ML pipelines

A data governance gate that enforces upstream integrity for AI/ML systems consuming open data (Socrata Open Data API / NYC PLUTO). It automates pre-ingestion schema validation, completeness checks, and distribution drift detection to block tainted features before any model consumes them.

  • Pre-ingestion schema and data-contract enforcement
  • Blocks drifted or incomplete features upstream
  • Immutable JSONL audit trail for data provenance

Tech stack

PythonsodapyRequestsPandasNumPy / PySparkGreat ExpectationsPydantic

Standards

NIST AI RMF (Govern & Map)EU AI Act Art. 10BCBS 239
Operational Risk · Impact04

Operations Impact Analysis

Algorithmic downstream impact & operational risk governance framework

A framework that looks past standalone model metrics to measure what model failures cost the business. It quantifies operational disruption, SLA degradation and organizational risk from model failures and edge cases, formalizing post-deployment impact assessment.

  • Scenario stress-testing of automated decision workflows
  • Sensitivity and bottleneck analysis for SLA risk
  • YAML/JSON operational risk registries for executives

Tech stack

PythonPandasNumPySciPyPlotlyMatplotlibStreamlit

Standards

NIST AI RMF (Manage & Measure)ISO/IEC 42001BCBS Op. Resilience
Regulatory Coverage

Frameworks I operationalize

How each project maps to the standards examiners and auditors ask about.

Framework01 · LLM Validation02 · ERM-X03 · QA Gate04 · Ops Impact
How I Work

Between policy, risk and engineering

Program management that turns requirements into systems teams can build and auditors can trust.

Policy → Controls

Requirements engineers can build

Decompose regulatory requirements into testable controls, owners and acceptance criteria that engineering teams can actually build against.

Evidence by Default

Compliance as a byproduct

Every pipeline emits structured, immutable audit artifacts, so compliance comes from running the system rather than a quarterly scramble.

Lifecycle Coverage

Data to decisions to impact

Governance spans the full chain — data entering the model, the model's decisions, and the operational impact those decisions have downstream.

AI Governance · Model Risk Management · Responsible AI Infrastructure

← Return to Website github.com/hibdiop