David Horvath

I have five years of professional experience, four of them building AI-powered products end to end — from training and deploying computer vision models and LLM/RAG systems to designing the distributed backends and cloud pipelines that turn them into shipped products.

At Verizon I trained custom YOLO models and served them from real-time video pipelines on AWS, built a VLM-powered security footage analyzer that triggered automated downstream workflows, and shipped one of the team's first production RAG systems. At Swiss Re I worked forward-deployed: embedded with business units as internal customers, building LLM-powered tools for investment analysis — including agent-based systems — alongside large-scale entity extraction pipelines in Palantir Foundry.

I'm now the software architect and sole engineer at Biomotion Technologies, where I designed the full stack of a shipping product — React/TypeScript frontend, C# backend, containerized Python services — and deploy specialized computer vision models as production inference services across a distributed multi-node system.

I like high autonomy and the parts of a product other people hand off: the device boundary, the offline update path, the architecture decision nobody wants to own. I spent a year as Scrum Master alongside development work, which mostly taught me how much of engineering is deciding what not to build.

Outside of product work I make interactive installations and live audio-reactive visuals — shown at Ars Electronica and the Zsolnay Light Festival, and collected in the Lab.