Applied AI and systems engineer.
I build AI systems and run them in production.
Every system on this page is running in production: acoustic bird identification on edge hardware at three family homes, a home-records app with durable AI workflows, and a private family archive with a multi-hop research agent, used by relatives who are not engineers. I designed, built, and operate each one solo, and every project page explains the decisions behind it.
Looking for applied AI or forward deployed engineering work, on contract through ToddTech LLC or full-time. Remote, or on site in southwest Michigan.
Portfolio
These aren't demos or mockups. Every system below is running in production right now, and each page explains what it does, how it was built, and the design decisions behind it.
Always-on acoustic bird identification at the edge: three family monitoring stations, one riding security-camera audio and two on budget lavalier mics, each feeding a living life list with AI-written field notes.
Runs privately on the family's own network; no public demo
View project details
The user manual your house never came with. Hearth turns photos, documents, and public records into the maintenance plan, inventory record, and historical archive your home has never had.
A private invite-only family archive with conversational AI search across decades of photos and documents, plus a multi-hop Family Researcher agent that follows the trail from one record to the next.
Capabilities
LLM systems
Retrieval-augmented chat with citations, a multi-hop agent that calls tools in a loop under a hard hop cap, structured output bound by schemas, and every model call routed through one gateway so models swap by config.
Proof Échoes, Hearth
Edge ML and computer vision
A custom-trained YOLO detector with multi-object tracking over live camera streams, 24/7 acoustic species classification with BirdNET, and inference pushed onto the smallest hardware that works: LiteRT on a no-AVX thin client, a Coral TPU in the NVR.
Proof The Aviary
Full-stack TypeScript
Next.js App Router with React Server Components and server actions, Supabase Postgres with row-level security as the real security boundary, durable background workflows, and realtime UI with a polling fallback.
Proof Hearth, Échoes
Enterprise .NET and legacy modernization
Ten years as project manager and lead developer, working remotely as a subcontractor to the prime, on the Naval Postgraduate School's Education Management System, a campus-wide application that had run on classic ASP since 1997. Led the team that rebuilt it in C#, ASP.NET MVC, and Entity Framework, launched on schedule in 2018, then built the API integrations that tied it to Active Directory, the campus business and research systems, and Power Automate and SharePoint through the Microsoft 365 transition. C#, ASP.NET MVC, Entity Framework, SQL Server, IIS, Windows Server, Active Directory.
Proof Naval Postgraduate School, remote subcontract, 2015 to 2025
Infrastructure and operations
Today, a six-machine Linux fleet on an MQTT event bus and a Tailscale mesh, systemd services with journald conventions, Docker, and stations at family homes that deploy themselves from a release branch. Before that, fifty-plus mixed Linux and Windows servers under Puppet configuration management and Nagios monitoring, with DoD-compliant patching and availability.
Proof The Aviary; Naval Postgraduate School, federal IT specialist, 2002 to 2015
Security
CompTIA Security+ (DoD 8140 IAT Level II), a decade as Information Assurance Officer in a Department of Defense environment, and data layers designed so the database enforces access rather than the UI.
Proof Every project
How I work
Issue-driven, docs in the same PR.
Every non-trivial change starts as a GitHub issue with an explicit contract and ends with a reviewed pull request that closes it. Each repo carries a living technical guide that is updated in the same commit set as the code. Main is always deployable, and merging is the deploy.
AI-assisted development with a system around it.
I use AI coding agents every day, and I built the discipline that makes them reliable: truth documents the agent reads before touching code, issue contracts with do-not-change lists, and a human review gate on every PR. Over 650 merged pull requests across three repos since April 2026.
Ship it, then keep it running.
Systems I build update themselves, fail loudly instead of silently, and run for weeks without a phone call. One monitoring station has run unattended at a relative's house since August 2026 on a fanless mini PC. Before that I ran a campus-wide enterprise application for ten years in a high-security government environment, through a full modernization launched on schedule and a decade of enhancements and integrations, without losing production stability.
Security is the starting point, not a feature.
Ten years of Information Assurance in a Department of Defense environment and an active Security+ certification. Access control, data privacy, and secure defaults are where the design begins.
I say when AI is the wrong tool.
Some problems need a spreadsheet, not a model. Where a model does belong, it gets the narrow job it is good at: in the Aviary, Python computes every statistic and the model only narrates them, so the prose can never invent a number.
I work next to the people who use it.
Ten years embedded with one customer, taking their requirements, building against their systems, deploying into their environment, and training their staff myself. Today the users of my archive are relatives who are not engineers and who tell me the moment something breaks. The language is whichever one the problem needs: I have shipped production systems in Perl, PHP, C#, Python, and TypeScript.
Todd Wyatt
I'm Todd Wyatt, an applied AI and systems engineer in Kalamazoo, Michigan, and a U.S. Navy veteran. I spent twenty-two years building for the Naval Postgraduate School: thirteen as a federal IT specialist, automating and securing a fleet of fifty-plus Linux and Windows servers, serving as Information Assurance Officer, and architecting ASP.NET applications; then, after moving home to Michigan in 2015, ten years as a remote subcontractor leading the team that rebuilt the campus-wide Education Management System, a classic ASP application dating to 1997, in C# and Entity Framework, and building the API integrations that connected it to Active Directory, Power Automate, and the rest of campus. Before that I was CTO of a small web firm in Carmel, building some of the first custom e-commerce platforms in the late nineties. In 2026 I started building AI systems end to end, from edge inference on thin clients to an agentic research assistant over a family archive, and running them in production for real people. CompTIA Security+ certified, and looking for the next team to build with.
Hiring or contracting?
Let's talk.
I'm open to contract and subcontract engagements through ToddTech LLC and to full-time roles, remote or in southwest Michigan. I've worked this way before: ten years as a remote subcontractor delivering and maintaining a federal enterprise system. Send a note about the role or the problem and I'll reply within one business day.
Contact Information
Name
Todd Wyatt
Location
Kalamazoo, Michigan
Response Time
Within 1 business day
Roles I'm targeting
- ›Forward deployed engineer / applied AI engineer
- ›ML engineer: edge inference, vision, audio
- ›Full-stack engineer: TypeScript, Next.js, Postgres
- ›Platform and infrastructure engineer
- ›.NET modernization and enterprise integration: C#, ASP.NET MVC, SQL Server, Active Directory, Power Automate
- ›Contract or subcontract engagements through ToddTech LLC
