Deepak Narayanan
Senior Applied Deep Learning Research Scientist at NVIDIA
Bellevue, Washington, United States
Summary
🤩
Rockstar🎓
Top SchoolDeepak Narayanan is a Senior Applied Deep Learning Research Scientist at NVIDIA with 13 years of experience building high-performance systems for training and serving ML models. He holds a PhD in Computer Science from Stanford and dual degrees from MIT, blending deep academic rigor with production-focused engineering. His work spans backend performance engineering and runtime systems—contributions include REPL and LLVM codegen improvements to the high-performance Weld runtime and runtime metrics enhancements for Stanford’s HELM evaluation framework. At Microsoft Research he bridged systems research and applied ML, and today he focuses on making large-model workflows more efficient at scale. Known for pragmatically refactoring complex code paths, he pairs strong systems instincts with a penchant for measurable, reproducible tooling.
13 years of coding experience
2 years of employment as a software developer
Bachelor's Degree, Mathematics and Computer Science, Bachelor's Degree, Mathematics and Computer Science at Massachusetts Institute of TechnologyNational Public SchoolDoctor of Philosophy (Ph.D.), Computer Science, Doctor of Philosophy (Ph.D.), Computer Science at Stanford University
Github Skills (30)
benchmark10
performance-monitor10
performance-analytics10
python10
benchmarking10
performance-measurement10
java10
javas10
performance-analysis10
language-modeling10
performance-tuning10
performance-monitoring10
llvm9
postgresql9
debug9
Programming languages (6)
Github contributions (5)
stanford-futuredata/macrobase
Dec 2015 - Mar 2016
MacroBase: A Search Engine for Fast Data
Role in this project:
Back-end Developer
Back-end DeveloperContributions:179 commits, 30 PRs, 193 pushes in 2 months
Contributions summary:Deepak primarily focused on fixing issues related to the loading of queries and server configuration within the `MacroBase` project. They also contributed by adding a new demo file that draws inliers and outliers from different normal distributions. Furthermore, they fixed various warnings and added timing for outlier detection and streaming classes, indicating a focus on performance and debugging. The changes suggest a strong understanding of the project's backend and its data processing logic.
dataframesmacrosquery-languagesearch-enginedatabase
weld-project/weld
Dec 2016 - Dec 2017
High-performance runtime for data analytics applications
Role in this project:Back-end Developer
Back-end DeveloperContributions:101 commits, 47 PRs, 106 pushes in 11 months
Contributions summary:Deepak primarily worked on the REPL (Read-Eval-Print Loop) for the Weld project, improving its functionality and usability. Their contributions included refactoring code, specifically cleaning up unwrapping logic, and adopting the `rustyline` package for a better REPL experience. Furthermore, the user implemented cast operations, including changes to parsing, type inference, and LLVM code generation, extending the Weld's capabilities. The user's efforts also resulted in LLVM code generation and the addition of tests for cast operations.
data-analyticsanalyticscode-generationdatastanford
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.