AI engineering · LLM, agentic & vision systems in production

David Horvath

I build AI systems that run in production — LLM pipelines, agentic workflows, and custom computer vision — and the distributed backends that carry them from prototype to shipped product.

Currently: software architect and sole engineer at a Vienna startup — designing the full stack and deploying containerized AI inference in production.

View work Get in touch

Selected work

Things that shipped

Three production systems, written up the way an engineer would explain them: the constraints, the decisions, and what runs unattended today.

HírSpektrum wordmark over the tagline “Nem hír. Narratíva.”

hirspektrum.hu

A self-hosted platform running continuous LLM pipelines in production: the Hungarian news feed, clustered without supervision, analyzed by orchestrated LLM workflows over a knowledge graph.

  • LLM pipelines
  • RabbitMQ
  • Neo4j
  • Clustering
  • Python
A neural network visualisation classifying a handwritten digit

Vision to LLM, in production

Four years of applied AI on AWS: custom YOLO models behind real-time video pipelines, a VLM-powered footage analyzer triggering automated workflows, production RAG, and forward-deployed LLM agents for financial analysis.

  • AWS
  • VLM
  • RAG
  • LLM agents
  • YOLO/PyTorch
A bioprinter nozzle extruding material into a culture vessel

Bio3D printer software

The full software architecture of a shipping product, owned by one engineer: distributed multi-node system, containerized CV inference in production, and a versioned release pipeline for air-gapped deployments.

  • System design
  • CV inference
  • C#
  • React/TS
  • CI/CD

The Lab

Where computer vision goes to dance

Interactive installations, live audio-reactive VJ sets, and real-time experiments. Shown at Ars Electronica and the Zsolnay Light Festival.

Enter the Lab

About

David Horvath

Five years building software, four of them building AI-powered products end to end — computer vision and production RAG at Verizon, forward-deployed LLM agent systems at Swiss Re, and now the full architecture of a shipping product, including its AI inference services.

I like the parts other people hand off: the deployment path, the release process, the architecture decision nobody wants to own.

More about me Download CV (PDF)

Working with: Python · TypeScript/React · C# · PyTorch · LLMs / RAG / agent systems · AWS · Palantir Foundry · Docker & Kubernetes · RabbitMQ · PostgreSQL / Neo4j · Jetson (GPU inference)