AI work I've shipped.
I've built AI into products and operations across defense research, fintech, luxury commerce, field services, and B2B SaaS. The thread connecting most of it: taking unstructured human input and turning it into structured business signal.
Some of this work I architected and built directly. Some I led with engineering and design teams. Some was AI layered into existing platforms with partners. All of it shipped to production and got used by real people for real work.
What I've built
Field input intelligence.
Built AI features that take unstructured technician input (handwritten notes, voice memos, job site photos) and convert them into structured business intelligence. Currently in production at a 400-person HVAC, electrical, and plumbing service business, where the work changed how the company sees what's happening across hundreds of jobs each day.
Multi-agent form building.
Designed and built a multi-agent AI system that compressed a five-day manual form-building process to same-day delivery for a B2B forms platform. The original schema had to be rebuilt because checkbox and radio button inconsistencies capped accuracy at 95%. After the schema redesign, the system held a 99% accuracy threshold with human-in-loop review for edge cases. The work unlocked healthcare and government markets the company couldn't previously serve.
Pricing intelligence for luxury markets.
Built an AI-powered pricing engine for a luxury watch retailer's trade-in and consignment workflows. The engine pulls from external market data and internal sales history to recommend pricing in a category where transaction values run high and pricing precision matters. Also built an AI brand voice system for product listings.
Defense research curation.
Built AI components for a digital curation system at the Air Force Research Laboratory through Optimal Efficiency. The work covered artifact deduplication, automatic tagging, similar-artifact recommendations, and summarization, designed to help researchers find and build on prior work. I built the underlying model components myself, which is rare for this category. Most AI consultants prompt and integrate. I've trained models.
Self-service business intelligence.
Set up a data lake and connected a natural-language BI tool on top of it for a service business operations layer. The team can now ask the system questions in plain language and get reports back. What used to be an engineering ticket is now a conversation.
Natural language search.
Built search systems that understand how humans actually ask questions. Used at a veterans benefits marketplace for member benefit discovery, and at the Air Force Research Laboratory for the digital curation system.