Avshalom Manevich

Research Engineer · Paris

About

I work at the intersection of research and production, making large multimodal models fast, efficient, and reliable to serve. I'm currently the inference team lead at H Company, where I build the serving stack behind the Holo vision-language models and the Surfer computer-use agents.

Before that, I worked on production LLM inference and the Jamba family of hybrid Transformer–Mamba models at AI21 Labs, and did research at Amazon's Alexa Shopping group and the Bosch Center for AI. My research at Bar-Ilan University's NLP lab focused on decoding-time methods for reducing hallucinations in vision-language models. Underneath it all is a decade of distributed-systems and data engineering.

Experience

Earlier Software Engineering

Education

Selected Publications & Writing