- 🔭 I enjoy building AI where the interesting part starts after the model works — when it has to handle real data, real scale, failures, GPUs, storage, deployment, and people actually using it.
- 🌱 Right now I'm focused on large-scale unsupervised visual detection and recognition, especially training and detection pipelines that can search through millions of images while keeping accuracy high and latency practical.
- 👁️ A lot of my recent work has been around visual intelligence systems: extracting strong embeddings on GPUs, organizing large image collections, searching them efficiently with vector indexes, and turning recognition results into something useful through desktop and real-time workflows.
- ⚙️ I care a lot about the engineering behind the model too. I've worked on resumable training, crash recovery, immutable vector-index generations, atomic data publication, and safe long-running pipelines so an interrupted job does not mean starting everything again — or exposing half-built state to the live system.
- 🐳 I also like taking ML all the way to deployment: offline runtimes, Docker, CUDA, compiled Python applications, release automation, licensing, Linux desktop integration, and reproducible environments. Making the system easy to ship is part of the engineering for me, not an afterthought.
- 🤖 Robotics is the other side of what I enjoy. I'm interested in combining perception, local LLMs, and hardware so robots can understand their environment and keep useful context over time.
- 🚀 My goal is simple: build AI systems that are not only impressive in a demo, but dependable enough to live in the real world.
- 💬 Ask me about AI, Robotics, Computer Vision, Deep Learning, RAG, LLMs, GPU inference, vector search, large-scale retrieval, Dockerized ML, or production AI systems.
I use different parts of this stack depending on the problem — these are tools and technologies I've actually worked with.
