Summary
Andrew Li is a software engineer and ML researcher with 12 years of experience building efficient computer vision and language models, currently reducing serving latency and memory for production Bard models at Google. He specializes in neural architecture search, mixed-precision quantization, and TPU-friendly model design—contributions that include EfficientNet-X and hybrid conv-transformer architectures with published work in CVPR and ASPLOS. His research blends practical production impact (significant speedups and latency reductions) with novel AutoML search and latency-aware scaling methods that find Pareto-optimal model families. A Berkeley-trained engineer, Andrew also has a track record of making heavy-weight vision tasks far more efficient (e.g., a 50x improvement in amodal segmentation) and brings a knack for turning theoretical insights into deployable optimizations. Outside of work he balances large-model efficiency with quieter obsessions—playing Go, singing, and camping—which often informs his iterative, strategic approach to ML design.
12 years of coding experience
3 years of employment as a software developer
Five Year Master of Science - MS, Electrical Engineering and Computer Science, Five Year Master of Science - MS, Electrical Engineering and Computer Science at University of California, Berkeley
Bachelor of Engineering (B.E.), Electrical Engineering and Computer Science, Bachelor of Engineering (B.E.), Electrical Engineering and Computer Science at UC Berkeley College of Engineering
English, Chinese