Skip to content

Projects

What I've Built

Shipped products, internal tools, open source, and side projects. Some of these solve real problems. Some of them just seemed like a good idea at the time.

Sentinel

Local repo health monitor with 14 detectors, an LLM judgment layer, and a web UI for triaging findings. Scans overnight, produces a morning report. The interesting part: cross-artifact analysis that catches docs drifting from code — something no existing tool does well.

PythonLLMCLIDeveloper Tools

GitHub →

TSGBuilder

Multi-agent pipeline that researches, writes, and reviews troubleshooting guides for Azure AI support engineers. Automated PII checks, structured quality gates. Cut guide authoring time by 70–80% and is rolling out across Microsoft support engineering.

Multi-AgentPythonAzure AIGitHub Copilot

GitHub →

WyoClear

Wyoming legislature tracker I built from scratch by directing AI agents across the full development lifecycle. Find legislators, track bills, see votes. No ads, no spin. Reached hundreds of users, expanding toward paid features.

PythonAI AgentsCivic TechFull Stack

Visit →

Write it Right

My LLC. Started as an AI-powered job application pipeline (cover letters, resumes tailored to specific roles), then evolved into consulting — helping small businesses build custom software and AI tools they couldn't afford before.

GoPythonLangChainConsulting

Visit →

Copilot Autonomous Template

A copier template that bootstraps a full autonomous AI development workflow — builder agent, reviewer, planner, tester subagents, stop hook, vision lock, ADRs, checkpoint protocol. Extracted from 20+ autonomous sessions building Sentinel.

CopilotAI WorkflowTemplateOpen Source

GitHub →

Rithm Digital Art

Digital artwork from simulating emergent natural behavior — physarum polycephalum, flocking, reaction-diffusion. Built in Unity3D. The kind of project that reminds you programming can be beautiful.

Unity3DGenerative ArtSimulation

View →

chrome-dino AI

AI that plays Chrome's dinosaur game. Recently rewritten with modern reinforcement learning (DQN) using PyTorch, Stable Baselines, and Tesseract OCR. The original 2018 version used a CNN trained on human input.

PyTorchRLDQNOCR

GitHub →