Skip to main content

Technology Leader in
Applied AI

Technologist in Residence at Nava PBC · Former U.S. Digital Service, Slalom, Ibotta, JazzHR

Technology leader in applied AI for benefits and public services, with 15 years of experience scaling teams and shipping production systems. Technologist in Residence at Nava PBC, where I lead the Nava Labs prototyping squad. My research tests AI against ground truth that can be checked — benefit formulas defined in statute, small models that run on the device, and evaluations with gates set before the runs. Previously led engineering at Department of Veterans Affairs Office of the CTO (USDS) and high-growth startups.

Notable Companies & Features

Logo of the United States Department of Veterans Affairs
company logo of johnson and johnson
company logo of slalom consulting
company logo of Ibotta
company logo of JazzHR
company logo of SlingTV
company logo of Vietnam Backpacker Hostel
Business Insider Logo
Guinness World Records Logo
Mashable Logo
MTV Australia Logo
New York Times Logo
Logo of the United States Department of Veterans Affairs
company logo of johnson and johnson
company logo of slalom consulting
company logo of Ibotta
company logo of JazzHR
company logo of SlingTV
company logo of Vietnam Backpacker Hostel
Business Insider Logo
Guinness World Records Logo
Mashable Logo
MTV Australia Logo
New York Times Logo
Logo of the United States Department of Veterans Affairs
company logo of johnson and johnson
company logo of slalom consulting
company logo of Ibotta
company logo of JazzHR
company logo of SlingTV
company logo of Vietnam Backpacker Hostel
Business Insider Logo
Guinness World Records Logo
Mashable Logo
MTV Australia Logo
New York Times Logo

Latest Posts

Thoughts on applied AI, engineering leadership, and building in the open

Research ai govtech +3

Where the Errors Actually Live

One hundred twenty-two experiments on SNAP payment error: a synthetic case generator with three validation gates, a data leak found by comparing against another lab's published results and corrected in public, and a black-box recovery of the benefit formula from real QC data. The formula layer checks out against the statute at 99.9%. The error lives in the fact layer: earnings that never reached the issued benefit.

Read more
Research ai govtech +3

Synthetic Data Is Only as Good as Its Provenance

I generate synthetic benefits cases from rules encoded out of primary sources and households sampled from real Census data, so every label is exact by construction. Verifying five years of those labels against the original FNS charts found two states that really did change policy in 2025, and a half-dozen parameters that had been quietly wrong. Synthetic data does not remove the hard problem. It moves it to source verification.

Read more
Research ai government +3

Weight-Space Programming for the Social Safety Net

An independent test of the claim that AI models can write, read, and edit neural network weights directly, run against US benefit formulas where the correct answer is defined by law. Written weights were exact. Trained weights failed in ways average accuracy hides. And one step of the law cannot be represented at all.

Read more

Research

Long-form work tested against ground truth that can be checked: benefit formulas defined in statute, synthetic data with exact labels, and payment records with coded errors.

All Research All Research link
Research ai govtech +3

Where the Errors Actually Live

One hundred twenty-two experiments on SNAP payment error: a synthetic case generator with three validation gates, a data leak found by comparing against another lab's published results and corrected in public, and a black-box recovery of the benefit formula from real QC data. The formula layer checks out against the statute at 99.9%. The error lives in the fact layer: earnings that never reached the issued benefit.

Read more
Research ai govtech +3

Synthetic Data Is Only as Good as Its Provenance

I generate synthetic benefits cases from rules encoded out of primary sources and households sampled from real Census data, so every label is exact by construction. Verifying five years of those labels against the original FNS charts found two states that really did change policy in 2025, and a half-dozen parameters that had been quietly wrong. Synthetic data does not remove the hard problem. It moves it to source verification.

Read more
Research ai government +3

Weight-Space Programming for the Social Safety Net

An independent test of the claim that AI models can write, read, and edit neural network weights directly, run against US benefit formulas where the correct answer is defined by law. Written weights were exact. Trained weights failed in ways average accuracy hides. And one step of the law cannot be represented at all.

Read more
Research ai privacy +2

Nine Versions of a Redaction Model

I trained my first model to redact PII and PHI on-device: an 11-million-parameter token classifier that fits in 11.3 MB. Nine versions of synthetic data, strict span scoring, and a failure ledger got it to 0.974 full-value recall on my golden set, with the misses still named in the post.

Read more

Featured Projects

A selection of technical and leadership projects that showcase my expertise

View All Past Projects

Technical Projects

Technical Project

phi-redact

An 11M-parameter model that finds PHI and PII - patient names, MRNs, dates, phone numbers and 26 other types - in English text, small enough to run in the browser so the text never leaves the device. Roughly tied with Desert Ant's Redact on English recall at half the size.

PyTorch Knowledge Distillation ONNX Runtime Web WebAssembly Astro
View Project
Technical Project

Agentic Auth

How NIST's IAL/AAL framework applies when AI agents act on your behalf - prompt injection, dry runs, audit trails, and what agent-aware systems actually need to build.

Astro NIST AI Agents
View Project
Technical Project

Snap Sentiment

Automated & Verified Reviews for Your Business - helping companies collect and display verified customer reviews.

Next.js Stripe AWS Tailwind
View Project
Technical Project

Email Bits

Beautiful HTML Email Designs Made Easy - a CodePen-inspired platform specifically for creating, editing, and sharing email templates.

Machine Learning React Tailwind AWS
View Project
Technical Project

In Short Pod

Personalized Podcast platform built during Buildspace Season 5 - converting your interests, location, and more into a personalized audio experience.

Next.js Generative AI Text To Speech AWS Buildspace
View Project
Technical Project

Cabeça de Queijo

Build a bi-lingual fan club website for the Green Bay Packers fans in Brazil. Figuring out how to grow the fan club and get more people involved.

Astro AWS i8n
View Project

Leadership & Innovation

For Fun

AI Systems Advisory

I advise teams building agentic and LLM-powered systems: architecture, tool and orchestration design, and evaluation. Agent orchestration on AWS Bedrock AgentCore, Model Context Protocol server design, and retrieval and evaluation architecture for compliance-sensitive domains.

Architecture reviewAdvisoryHands-on consulting