Devashish Shankar
Staff Software Engineer at Meta
Sunnyvale, California, United States
Summary
🤩
Rockstar🎓
Top SchoolDevashish Shankar is a Staff Software Engineer with 14 years of experience building production-grade ML and data systems, currently based in Sunnyvale and working at Meta. He progressed from SDE roles at Flipkart to founding technical positions at Drishti, where he led architecture and data-science initiatives before moving into senior ML engineering at Meta. Devashish combines deep systems and ML fluency with practical compiler-level optimizations, exemplified by significant contributions to PyTorch’s Inductor pattern matcher to optimize split-cat scenarios and graph compilation. He’s comfortable moving between low-level performance work and high-level product requirements, delivering both scalability and inference efficiency. His career blends startup-founders’ breadth with large-company rigor, making him effective at translating research-grade improvements into production impact. Trained in computer science at VIT, he brings a pragmatic, performance-first mindset that often surfaces non-obvious compiler and graph-level fixes to speed ML workloads.
14 years of coding experience
11 years of employment as a software developer
Bachelor of Technology (B.Tech.) Computer Science, Bachelor of Technology (B.Tech.) Computer Science at Vellore Institute of Technology
English, Hindi
Github Skills (12)
compiler-optimization10
pytorch10
machine-learning10
pattern-matching10
python10
autograd9
neural-network9
deep-learning9
tensor9
numpy8
gpu8
deeplearning-ai7
Programming languages (5)
Github contributions (5)
pytorch/pytorch
Mar 2023 - Aug 2024
Tensors and Dynamic neural networks in Python with strong GPU acceleration
Role in this project:
ML Engineer
ML EngineerContributions:27 reviews, 22 PRs, 105 pushes in 1 year 4 months
Contributions summary:Devashish made significant contributions to the `pytorch/pytorch` repository, primarily focused on improving the pattern matcher used within the Inductor compiler. They extended the pattern matcher to handle more complex split-cat scenarios, optimizing the graph compilation process. Key changes include adding new pattern matching capabilities, handling multi-user cases, simplifying split-cat patterns, and replacing squeeze operations with unbind/stack operations. This work directly impacts the performance and efficiency of PyTorch's compilation pipeline.
pythongpu-accelerationdeep-learninggpunumpy
devashishshankar/pytorch
Mar 2023 - Aug 2024
Tensors and Dynamic neural networks in Python with strong GPU acceleration
Contributions:73 pushes, 13 branches in 1 year 4 months
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.