Sandeep Subramanian
Research Scientist at Mistral AI
San Francisco, California, United States
Summary
🤩
Rockstar🎓
Top SchoolSandeep Subramanian is a research scientist with 12 years of experience specializing in deep learning and NLP, currently at Mistral AI after research and engineering roles at NVIDIA and top academic labs like MILA and CMU. He holds a PhD from Université de Montréal and has interned at leading labs including Facebook AI and Microsoft Research, blending rigorous academic research with production-focused ML engineering. At NVIDIA he contributed to NeMo, enhancing transformer architectures and efficient data handling for large-scale generative and speech models—work that directly supports widely used open-source tooling for LLMs and multimodal research. His background spans sequence-to-sequence modeling, attention mechanisms, and core framework development, with a track record of integrating research advances into robust back-end systems. Based in San Francisco, he pairs leadership experience with hands-on implementation, often moving models from prototype to scalable training pipelines. A less obvious strength is his repeated pattern of contributing low-level model and data infrastructure improvements that unlock higher experimental throughput for research teams.
12 years of coding experience
5 years of employment as a software developer
Doctor of Philosophy (Ph.D.) Deep Learning, Doctor of Philosophy (Ph.D.) Deep Learning at Université de MontréalBachelor of Technology (B.Tech.) Computer Science and Engineering, Bachelor of Technology (B.Tech.) Computer Science and Engineering at Vellore Institute of TechnologyMaster's Degree NLP, Master's Degree NLP at Carnegie Mellon University
English, Tamil, Hindi
Github Skills (14)
attention-mechanism10
pytorch10
lstm10
rnn-model10
nlp10
deep-learning10
large-language-models10
python10
n10
seq2seq10
machine-translation9
data-handling8
generative-ai8
tensorflow7
Programming languages (4)
Github contributions (5)
MaximumEntropy/Seq2Seq-PyTorch
Jan 2017 - Jun 2017
Sequence to Sequence Models with PyTorch
Role in this project:
Backend Developer & ML Engineer
Backend Developer & ML EngineerContributions:29 commits, 2 PRs, 43 pushes in 4 months
Contributions summary:Sandeep implemented LSTM-based sequence-to-sequence models with attention mechanisms using PyTorch. Their contributions involved defining and modifying LSTM cells, incorporating attention mechanisms, and refactoring code for a new attention mechanism. The user also worked on the overall architecture of a sequence-to-sequence model, including embeddings and decoding functionalities, with a focus on sequence generation.
pytorchsequencernnseq2seqsequence-models
NVIDIA/NeMo
Nov 2020 - Jan 2023
A scalable generative AI framework built for researchers and developers working on Large Language Models, Multimodal, and Speech AI (Automatic Speech Recognition and Text-to-Speech)
Role in this project:Back-end Developer & ML Engineer
Back-end Developer & ML EngineerContributions:638 reviews, 826 commits, 324 PRs in 2 years 2 months
Contributions summary:Sandeep's contributions center around adding and modifying core components of the NLP framework. They implemented and integrated Pre-LN transformer encoder/decoder layers, indicating a focus on improving the architecture's performance. Furthermore, the user worked on adding support for tarred datasets for NMT, expanding the framework's data handling capabilities and enabling more efficient training. They also worked on fixing bugs within the model.
asrspeech-recognitionnatural-language-processingttsspeaker-diarization
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.