Rithesh Kumar
Member Of Technical Staff at OpenAI
San Francisco, California, United States
Summary
🤩
Rockstar🎓
Top SchoolRithesh Kumar is a speech AI researcher and engineer with 11 years of experience building production-grade voice generation systems, currently a Member of Technical Staff on OpenAI’s realtime speech team. He led speech generation research at Adobe, shipping multilingual, broadcast-quality diffusion models for voice translation and text-to-avatar features, and previously drove Descript’s Overdub and Regenerate products that deliver ultra-realistic voice cloning and edit-at-text workflows. His work spans scaling diffusion models, efficient distillation for multilingual audio, and latent-space modeling—reflected in open-source contributions such as a PyTorch VQ-VAE with PixelCNN improvements. Comfortable bridging research and product, he has a Mila-trained academic background and a track record of taking prototype generative models into widely used features. Notably, he pairs deep generative-model expertise with hands-on engineering that ships at 44.1 kHz production quality.
11 years of coding experience
8 years of employment as a software developer
Master of Science - MS Computer Science Track, Master of Science - MS Computer Science Track at Mila - Quebec Artificial Intelligence InstituteAnna University, ChennaiHigh School Computer Science, High School Computer Science at Bala Vidya Mandir
Github Skills (5)
pytorch10
deeplearning-ai10
deep-learning10
python10
cifar1009
Programming languages (3)
Github contributions (5)
ritheshkumar95/pytorch-vqvae
Apr 2018 - May 2018
Vector Quantized VAEs - PyTorch Implementation
Role in this project:
ML Engineer
ML EngineerContributions:20 commits, 5 PRs, 17 pushes in 1 month
Contributions summary:Rithesh contributed to the implementation of a Vector Quantized Variational Autoencoder (VQ-VAE) using PyTorch. Their work included adding CIFAR-10 data loading and training code within the `main.py` file. Further improvements involved refactoring the code and introducing a PixelCNN model for latent space modeling, and then shifting toward the Gated PixelCNN to improve generative capabilities.
pytorchdeep-learningvaegenerative-modelsvq-vae
ritheshkumar95/pytorch_handwriting_synthesis
Jan 2019 - Mar 2019
Handwriting Synthesis and Prediction - PyTorch Implementation
Contributions:36 commits, 1 push in 2 months
pytorchgenerative-modelshandwriting-generationrnnsseq2seq
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.