Summary
Sudhanshu Agrawal is a Machine Learning Research Engineer at Qualcomm AI in San Diego with nine years of hands-on experience building and optimizing generative models for on-device inference. He holds dual BS degrees in Computer Science and Pure Mathematics from UCLA and has contributed to NeurIPS workshops on speculative decoding and adaptive algorithms that boost LLM inference speed. His work spans transformers, diffusion models, KV caching, LoRA, and system-level tooling—recently delivering record on-device inference performance and a novel adaptive draft length speculative decoding algorithm. Prior research at UCLA and publications in computational physics demonstrate a strong theoretical foundation paired with practical implementation skills in Python, Julia, and distributed systems. Notably, he built a Python profiler adopted internally at Qualcomm and synthesized 100k+ webpages to solve a data-scarcity problem for CV training, reflecting a pattern of creative engineering solutions beyond core ML modeling.
9 years of coding experience
5 years of employment as a software developer
Indian School Certificate, Physics, Chemistry, Mathematics, Computer Science, English, National Rank 4 in the ISC Board (99.4%), Indian School Certificate, Physics, Chemistry, Mathematics, Computer Science, English, National Rank 4 in the ISC Board (99.4%) at Mallya Aditi International School
Indian Certificate of Secondary Education, Physics, Chemistry, Biology, Maths, Computer Science, English, History-Civics, Geography, Hindi, School Rank 1 (97.0%), Indian Certificate of Secondary Education, Physics, Chemistry, Biology, Maths, Computer Science, English, History-Civics, Geography, Hindi, School Rank 1 (97.0%) at Sishu Griha Montessori And High School
Bachelor of Science - BS, Mathematics and Computer Science, 3.92, Bachelor of Science - BS, Mathematics and Computer Science, 3.92 at University of California, Los Angeles
English, Hindi