Kevin Williams

Kevin Williams

Engineer & AI Researcher

I find problems worth solving and build what's needed to solve them. Systems engineer and AI researcher, currently building Safety Critical Labs and working on human spaceflight systems at NASA JSC.

GitHub ORCID Resume Email

Experience
Jan 2026 — present · Oct 2023 — Feb 2025

Systems Engineer

Barrios Technology → NASA Johnson Space Center

Developing, reviewing, and verifying systems engineering requirements across ISS, Orion, and the Commercial Lunar Delivery Program (CLDP), including interface and safety requirements for commercial partner integration. Integrated AI requirements framework into CLDP, establishing evaluation criteria for autonomous systems in safety-critical human spaceflight environments. Defining data architecture for the Extravehicular Activity and Human Surface Mobility Program (EHP). Coordinating with commercial partners including Axiom Space on requirements alignment and NASA human spaceflight safety compliance.

NPR 7123 NPR 7150.2 MBSE V&V CLDP
Feb 2025 — Dec 2025

Software Engineer

Barrios Technology

Architected and deployed Cortex, a unified enterprise data platform with integrated LLM capabilities for Intuitive Machines, replacing fragmented workflows with a single interface for document management, data retrieval, and cross-team collaboration. Developed cross-platform Flutter application for KULR Technologies automating report generation and operational workflows.

Django Python Flutter LLMs
Jun 2022 — Sep 2023

Engineering Specialist

Ford Motor Company

Led 20-person eMotor materials laboratory — metallurgical cross-sectioning, microscopy, hardness, and tensile testing to characterize materials and qualify components for electrified powertrain programs. Conducted failure analysis and root cause investigations, contributing to DFMEA/PFMEA processes. Designed test fixtures and measurement systems using CAD with GD&T. Developed automated inspection systems feeding quantitative data back into design reviews.

Materials Deep Learning Failure Analysis CAD/GD&T
Sep 2019 — Jun 2022

Research Assistant

University of Michigan

Developed novel electrochemical detection method for Δ9-THC (1–20 μM) achieving 0.13 μM limit of detection with R² = 0.995, third-lowest detection limit among comparable SPCE devices. Published peer-reviewed findings demonstrating viability for in-field applications.

Electrochemistry Biosensing Publication
2012 — 2017

Materials & Chemical Technician

Hyundai-Kia · General Motors · Quaker Chemicals

Metallurgical analysis, material validation, and quality testing across automotive OEMs. Established GM Flint Engine metallurgical lab. Led supplier corrective actions at Hyundai-Kia. Trained personnel on QA protocols.


Education

M.S. Artificial Intelligence

University of Michigan · May 2026

B.S. Mechanical Engineering

University of Michigan · 2021 · Featured in U-M ME150

Research Projects

AI Requirements Framework

Open standard for certifying AI in safety-critical systems. Nine requirements (AI-1 through AI-9) covering bias, validation, hallucination, and human-AI teaming. Gap analysis confirmed no existing standard adequately addresses AI-specific failure modes in human spaceflight.

Privacy-Preserving Cognitive Twin

Adversarial evaluation pipeline studying model vulnerability to membership inference attacks — confidence-based, label-only, and shadow model variants. Benchmarked three defense strategies: DP-SGD via Opacus, Model Confidence Exclusion, and a fusion defense, sweeping ε values to empirically map privacy–utility tradeoffs.

μStruct — AI Microstructure Identification

Ensemble classifier combining CLIP zero-shot inference with FAISS similarity-weighted voting across 6,000+ micrographs and 16 steel microstructural phases. Productionized with FastAPI and Docker on Hugging Face Spaces.

Disaster Infrastructure Damage Recognition

CNN-based binary classifier for infrastructure damage assessment in post-event aerial and ground imagery. Empirical evaluation of classification confidence thresholds for first-responder deployment readiness.


Publications
Requirements and Verification Standards for Artificial Intelligence in Safety-Critical Applications
Kevin Williams · Safety Critical Labs · Zenodo, 2026
Differential pulsed voltammetry of Δ9-THC on disposable screen-printed carbon electrodes: A potential in-field method to detect Δ9-THC in saliva
2023 · University of Michigan

Awards
NASA Silver Bear Award — Certification of AI on Human-Rated Space Flight Systems
Ford Technical Excellence Award — Deep Learning for Quality Decisions

Beyond the Code

I think a lot about what it means to live well — love, kindness, integrity, the discipline of showing up when you don't feel like it. These aren't separate from the engineering. They're the foundation of it. The way I approach AI safety starts with the same questions philosophy has always asked: what do we owe each other, and what does it look like to build systems that reflect that?

I train every day, produce music, and still want to be an astronaut. I don't have social media. I'd rather be known for what I build and how I treat people than for a curated presence.