Mahesh Doijade
Sr. Compute Developer Technology Engineer at NVIDIA
Pune, Maharashtra, India
Summary
🤩
Rockstar🎓
Top SchoolMahesh Doijade is a senior compute developer with 8+ years of hands-on experience accelerating ML, DL and scientific workloads on GPUs at NVIDIA, where he currently focuses on LLM inference and high-performance compute. He brings deep expertise in C/C++, CUDA, CUTLASS, NVSHMEM, OpenCL, MPI and OpenMP, and has a track record of improving core RAPIDS/RAFT primitives—adding numerous distance metrics and optimizing fused L2 kNN kernels for better compute/memory overlap. Earlier work includes CUDA SDK sample ownership and porting performance features across CUDA toolkits, plus OpenCL/CUDA acceleration for x265, reflecting a strong applied systems background. Notably, his contributions removed cutlass dependencies in cuML and introduced efficient contraction kernels, demonstrating an ability to simplify complex GPU stacks while boosting performance. Based in Pune, he pairs academic HPC roots (parallel SAT solver, DRDO-sponsored projects) with production-grade GPU engineering for ML and scientific applications.
8 years of coding experience
8 years of employment as a software developer
B.E., Information Technology, B.E., Information Technology at University of MumbaiMTech, Computer Science and Engineering, MTech, Computer Science and Engineering at Walchand College of Engineering
Github Skills (24)
algorithm10
vector-search10
distance-matrix10
c-language10
gpu-programming10
machine-learning10
machine-learning-algorithms10
rapids10
parallel-computing10
gpu10
cuda10
cuml10
cprogramming-language10
metric10
matrix-multiplication9
Programming languages (4)
Github contributions (5)
NVIDIA/cuda-samples
Mar 2018 - Mar 2021
Samples for CUDA Developers which demonstrates features in CUDA Toolkit
Role in this project:
Back-end Developer
Back-end DeveloperContributions:11 releases, 30 commits, 13 pushes in 3 years
Contributions summary:Mahesh made several commits related to the CUDA Toolkit samples. These commits focused on adding and updating sample code to support newer CUDA versions and features, including 10.0, 10.1, 10.1 Update 1, 10.1 Update 2, 10.2, 11.0, 11.1, and 11.2. The changes involved modifications to existing code and the addition of new samples demonstrating features such as matrix multiplication, bandwidth testing, and support for features like compressible memory, and other advanced CUDA APIs. The user also addressed build issues and updated documentation.
cudagpucuda-toolkitcuda-openglcuda-kernels
rapidsai/raft
Apr 2021 - Jan 2023
RAFT contains fundamental widely-used algorithms and primitives for machine learning and information retrieval. The algorithms are CUDA-accelerated and form building blocks for more easily writing high performance applications.
Role in this project:ML Engineer
ML EngineerContributions:38 reviews, 34 commits, 20 PRs in 1 year 10 months
Contributions summary:Mahesh primarily contributed to the development and optimization of fundamental algorithms and primitives for machine learning and information retrieval within the RAPIDS ecosystem. Their work involved implementing new distance metrics (L1, L2, cosine, Chebyshev, Canberra, Minkowski, Hellinger, Hamming, Jensen-Shannon, KL-Divergence, Russell-Rao, and Correlation) and enhancing the fused L2 kNN kernel for improved performance. Furthermore, the user focused on optimizing performance by overlapping compute and memory operations and adding specialized instructions for vectorized processing. The user's contributions also extended to fixing bugs and improving stability.
sciencedata-sciencemachine-learninggraph-data-sciencefundamental
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.