AI engineer · LLM systems, computer vision, production ML
I build AI that ships.
The AI I build does a real job. It reads long documents so you don't have to, and watches the cameras you can't. I also build the software around it, so it works for real people.
See the workMost models die in a notebook.
I build everything around them that lets them survive real traffic and real people.
I build what lets them survive real traffic.
Selected work
Each one shipped to a real user, or was built to find out if it should.
The rest, laid out on a canvas you can drag, or as a list.
What I'm made of
Five layers, from the model to the product around it, each with the work that proves it.
Build AI features
Chatbots, document readers and agents that look things up and check their own answers before they reach your users.
Tools: LangGraph, LangChain, RAG, pgvector, Pydantic
Data-to-Policy: long reports turned into short briefsMachine learning & deep learning
Models trained on your data, then turned into a service your app can call. Quick to answer, steady when traffic spikes.
Tools: PyTorch, scikit-learn, FastAPI, Celery, Docker
Warden: content checks that never keep anyone waitingComputer vision
Software that finds things in photos and video, and can show you why it made the call.
Tools: PyTorch, Ultralytics YOLO, EfficientNet, OpenCV
Gateworx: cameras that spot theft, fire and crowdsSoftware development
The website, the database and the deploy around the model, so it reaches people and doesn't stay a demo.
Tools: Next.js, Django, PostgreSQL, Kubernetes
Nexus: a platform where developers find teammatesData work
Experiments and analysis that end in a clear decision, with the doubt shown instead of hidden.
Tools: SQL, pandas, A/B testing, scikit-learn
Levant Yarn: ad experiments that guided spend