Summary
Soham Ghosh is an AI research scientist and software engineer with a decade of experience building large-scale vision and multimodal models, currently contributing to Mistral AI after roles at Google DeepMind and Google Cloud AI. He blends research rigor—having worked on SLAM and meta-reinforcement learning in Ruslan Salakhutdinov’s lab—with product-focused engineering, shipping video understanding and vision-language systems at cloud scale. His background spans startups to enterprise, including founding a data-driven insure-tech platform and internships at Amazon Alexa and IBM Research, giving him a rare mix of full-stack system building and ML research. Known for improving vision encoders via self-supervision and contrastive learning, he focuses on making visual understanding more scalable and semantically grounded. Based in Sunnyvale, he combines academic training from Carnegie Mellon and NTU with hands-on deployment experience, and maintains an interest in large models and multimodal RL that guide his research direction.
10 years of coding experience
8 years of employment as a software developer
Exchange Program Computer Science, Exchange Program Computer Science at University of California, Berkeley
GCE 'A' Level, GCE 'A' Level at Raffles Institution
Master of Science (M.S.) Computational Data Science, Master of Science (M.S.) Computational Data Science at Carnegie Mellon University
Bachelor of Engineering (BEng) Renaissance Engineering Programme, Bachelor of Engineering (BEng) Renaissance Engineering Programme at Nanyang Technological University Singapore
English, Bengali, Hindi