James R. A. Henry
James leads Eigan’s research and core technology. He has spent more than twenty-five years in internet service provider engineering and security, first as a systems architect at a Canadian cable provider and then in telecommunications cyber threat intelligence. That work comes down to one question: what is the system actually doing, as opposed to what it reports?
Since August 2025 he has applied that question to language models. He is the sole author of three papers on how concepts form across the depth of transformer models and how to extract stable probes from them, and of the open reference implementation, concept-pair corpus and activation data they rest on. He is largely self-taught in machine learning research.