I build AI systems and stress-test them. Security research on large language models, applied AI for a funded startup, and the games and tools in between.
Most people pick a lane. I work where security, AI and real-time systems overlap, and I go deep in all four.
LLM red-teaming and prompt-injection defense. Seraph Guard, hardened against the OWASP LLM Top 10 and tested against a live attacker.
Generative pipelines and local inference for voice, image and audio. Real tools that run, not slideware.
Real-time worlds in Unreal and Unity. Co-founder of Infinite Horizon, EXIST-funded AR learning.
3D, materials and procedural tools in Blender and Python. I make the assets I use.
A local guard layer that sits in front of an LLM and is hardened against the full prompt-injection kill-chain. Every defense is measured against a live red-team and published as a reproducible benchmark and a bilingual whitepaper.
Visit the research site ↗EXIST-funded AR climate learning for kids 8 to 12. Co-founded with a team of four; I own the AI systems and core development.
infinite-horizon.eu ↗A local multi-track audio studio. Voice, music and SFX generated and composed on one timeline, fully offline.
Read the deep dive ↗A node-based procedural planet texture generator with a real-time OpenGL preview. Exports PBR basecolor, height, normal and roughness maps for game and visualization pipelines.
Read the deep dive ↗A bit of the 3D, asset and material work I make in Blender and Substance. The artist side is part of how I build everything else.
Open for work in AI engineering and LLM security, research collaborations, and conversations about building AI products that hold up under attack.
