Abhishek Agrawal
Software Developer
I am a Software Engineer at Google, specializing in Machine Learning Infrastructure and Distributed Systems. Currently, I am a core contributor to the open-source Orbax checkpointing ecosystem, where my work focuses on performance scaling, format compatibility, and system reliability for large-scale ML frameworks. My recent initiatives center on optimizing infrastructure for massive Large Language Models (LLMs). I recently architected Safetensors loading optimizations that drastically reduced 1TB model load times from two hours down to approximately four minutes. Additionally, I have driven efforts to expand open-source interoperability by engineering robust APIs to natively integrate PyTorch and NumPy checkpoint support. Prior to Google, I built high-throughput prediction pipelines for real-time bidding systems at Media.net(Directi) and developed smart contracts at Samsung R&D. My approach to systems architecture is grounded in a strong algorithmic foundation, highlighted by my standing as a Codeforces Candidate Master, a LeetCode Knight, and an ICPC Regional Finalist. I am passionate about advancing open-source ML infrastructure and look forward to sharing scalable engineering solutions with the SFSCON community.
