Projects across many fields, built the same way.
Each project starts with learning how the work really runs. Every entry lists its real status, so you can see what is in daily use, what has shipped to a client, and what is still a prototype.
AutoTagger for Revit
A Revit solution that tags and numbers hundreds of doors and rooms in seconds for multi-unit buildings, built around the way one architecture firm already works.
Read the storyFunctionable
A private, installable daily companion that helps people with ADHD and anxiety start important work, practice coping skills in the moment, and plan tomorrow the night before.
Read the storyLiving kidney donation research
Seven years of studying one hard problem deeply, which produced methods for expert-labeled data, prompt optimization, multi-agent classification, and medical test sequencing that generalize well beyond it.
Read the storyFrom three marketing pitches to an owner who runs his own AI
The owner of a small sensor manufacturer avoided a marketing retainer of about $21,600 a year by learning to improve his own website with AI in a couple of hours.
Read the storySystem Studio
A prototype that turns what a team knows about its process into a shared model with sourced numbers, then matches that model to 41 well-studied problems from operations research.
Read the storyFrom research, government, and the plant floor
These projects come from roles before and alongside Phlox Engineering. They show the same habit of learning a field first and then building for it.
Material delivery redesign at Yanfeng Automotive Interiors
As a continuous improvement intern at a Louisville plant, I designed and implemented a streamlined material delivery system that reduced cost by $600,000, and presented the results to corporate leaders.
An AI review solution at the CDC
At the CDC's injury center, I built a review application that let experts compare their judgments with a language model's across nine criteria, and I taught nontechnical colleagues what the AI was doing and why. It turned prompt refinement into a measured feedback loop.
AI-assisted labeling workbench
At Houston Methodist, I built a workbench where a clinical team defines its criteria, watches an AI classify examples, and refines the instructions until the model matches expert judgment. The pilot won a $50,000 award.
Work order automation
I studied how facilities work orders flow, built an agentic prototype for the steps with the most friction, and presented the concept at an International Facility Management Association conference.
Tell me about the part of your work that makes you say “There's got to be a better way to do this!”
The first conversation is free and takes thirty minutes. You describe the work, and I tell you honestly whether I can help and what a first step would look like.