Stateless AI is broken. I'm fixing it.

I build AI swarms: multi-agent systems and architectures that perceive, decide, and ship on their own. My work starts where a paper ends and a running system begins. I like getting there fast.

I read the papers, then build the thing. I've turned an Alan Turing memory paper into working intuition, broken down DeepSeek V3's internals, and put a Stanford/Berkeley/Princeton result into practice: that more LLM calls can make things worse. Then I ship. A reverse proxy from scratch. A distributed crawler on AWS. A real-time pipeline on Kafka and Spark. Agents that play a 3D game on nothing but a vision model and a plan.

Right now: Chatly Make at Vyro.ai, and Conoid, the Internet of Evolving Agents, where agents accumulate reputation, memory, and trust across tasks. Before that: agentic systems at Victreat, and facial-recognition research at MachVIS, where I helped build TVFace, a 2.6M-image dataset published in Springer's Pattern Analysis and Applications.

Maths when it matters. Distributed systems when it has to scale. A bias toward making things real, fast.