Work
Mostly machine learning and the engineering around it, with a long detour through mobile. Each of these is here because something actually happened: it shipped, it won something, or it taught me the thing I needed next.
Roles
Where I've worked
Aug 2026 — present · Remote
The reason I left UnitedHealthcare, delivered through Vantix. Umbrage is a Bain & Company subsidiary; I lead AI engineering across their client engagements, embedding with the teams that own a business process to work out which part of it genuinely benefits from a model and which part just needs better software, then shipping the result.
Jan 2026 — present · Self-employed
A firm of forward deployed engineers, built on the belief that the value of AI shows up in production rather than in a proof of concept. We embed as an extension of a client's engineering org, target under six weeks from kickoff to deployment, and hand over full IP: production code, not recommendations. The work spans AI orchestration (RAG, agents, LLM ops), embedded data engineering, and post-acquisition tech-stack consolidation. Full time since August 2026, starting with Umbrage, a Bain & Company subsidiary.
May 2026 — Aug 2026 · Minnetonka, MN
Picked as one of 15 engineers out of UnitedHealth Group Technology's 13,000 to start the company's forward deployed engineering practice, embedded with business units rather than sitting on a platform team. Redefined the risk adjustment model inside the finance division, a system tied to more than $150B in revenue, and built an AI-backed vendor relations dashboard that let executives manage third-party relationships directly. Also built the FDE deployment platform itself, standardizing environments across AWS, Azure, and GCP so a new engagement could reach production without rebuilding the runway each time.
Jul 2025 — May 2026 · Minnetonka, MN
Architected an enterprise AI observability platform covering both legacy and cloud-native applications, preventing over $1M in annual losses by cutting mean time to recovery on systems where downtime is valued at $600K per minute. Also built custom pipeline tooling and MCP servers that let autonomous agents map complex system dependencies, and agent-driven data retrieval that made member information far easier to reach internally.
Jun 2024 — Jun 2025 · Eden Prairie, MN
Containerized 52 .NET APIs and 10 Angular web apps with Docker, Kubernetes, and Helm, and built the CI/CD pipelines behind them with GitHub Actions and Octopus Deploy. Used generative AI to automate the bash scripting that cloud-migration standards required.
Jun 2023 — Aug 2023 · Eden Prairie, MN
Designed CNN architectures for a computer-vision system that detects section headers in medical documents at a 90% success rate, using image processing, transfer learning, and character recognition to make the model robust. It shipped into a platform that has since processed more than 2M documents.
May 2021 — Sep 2022 · St. Louis Park, MN
Built client Android applications in Kotlin and Java across more than three apps: authentication, API integration, Google Maps, and the user interfaces on top of them.
Sep 2020 — May 2021 · Eden Prairie, MN
Volunteered at Chamber events and built connections with local businesses. The first thing on this list, and the only one with no code in it.
Projects
The built things
2026
A SaaS platform that finds the people worth knowing in a network you already have.
Surfaces professional connections by shared criteria (alumni, past employers, location) with a dashboard for matches, companies, and roles, AI-generated outreach emails, and a discovery feed that keeps suggesting new people.
2025
Document-layout computer vision, published on PyPI as an open-source library.
Makes document-layout CV and OCR accessible from a few lines of Python. An object-detection CNN trained on IBM's PubLayNet dataset parses the structure of a page, then Tesseract extracts the text region by region rather than treating the page as one flat block.
2023
Object detection plus inpainting, trained end to end to erase a moving watermark.
A YOLOv7 detector locates the watermark frame by frame, since it drifts around the video and a fixed mask will not work. Computer-vision inpainting then reconstructs what was behind it. The interesting part was the training data: the detector needed watermarks composited over enough backgrounds that it learned the mark rather than the content underneath.
2021
Won the Congressional App Challenge for Minnesota's 3rd district.
A Flask site that scraped and republished current COVID-19 figures during the period when official dashboards were slow and hard to read. Data came from the COVID Tracking Project API plus BeautifulSoup where the API fell short, running on a self-managed Ubuntu box.
2021
Shipped map, auth, and profile features into a nonprofit app with a global user base.
Worked on the Android client for Little Free Library: the Google Maps integration for finding nearby libraries, the authentication flow, profile pictures and user information, and several of the UI screens.
2023
A population that starts at random weights and reliably solves the game.
Reimplemented Flappy Bird so a genetic algorithm could drive it, then evolved a population of networks from random initialisation through selection and mutation until the birds cleared pipes indefinitely. Written up as a short paper.
2021
Released on the App Store with Game Center leaderboards.
A 3D iOS game built on SceneKit with models made in Blender, plus a Game Center leaderboard so players could compete. My first end-to-end ship: build, submit, review, release.
Tools
Grouped by use
Grouped by the kind of problem rather than listed as a wall of logos. Depth varies: the machine learning and backend rows are where I spend most of my time.
Education
Two lines