Karthik Prasad
Research Engineer, LLM (Llama) Research And Foundation
California, United States
Summary
🤩
Rockstar🎓
Top SchoolKarthik Prasad is an experienced ML researcher and engineer with a decade of hands-on work building and shipping privacy-aware and efficient systems for large-scale NLP and LLMs, currently driving mid- and post-training research for Llama at Meta. His background spans on-device ML, model compression, differential privacy and federated learning, and earlier roles tackled query intent, embeddings, and content moderation for Facebook Search. He has contributed to prominent open-source projects like PyText and Opacus—adding differential privacy support, per-sample gradients and convolutional-layer improvements—demonstrating a rare mix of research depth and production-grade engineering. Based in California, he pairs an academic foundation from UC Irvine with practical systems experience from Akamai and Adobe, and has quietly bridged neural-interface research (decoding EMG handwriting) into broader AR/VR input modalities.
10 years of coding experience
9 years of employment as a software developer
Bachelor of Engineering (B.E.) Computer Science and Engineering, Bachelor of Engineering (B.E.) Computer Science and Engineering at PES UniversityComputer Science Mathematics Physics Chemistry, Computer Science Mathematics Physics Chemistry at Vidya Mandir Independent P.U. CollegeUniversity of California, Irvine
Sanskrit, Kannada, Hindi, English
Github Skills (11)
neural-network10
pytorch10
machine-learning10
convolutional-neural-networks10
nlp10
differential-privacy10
python10
ml9
back-end-development9
deep-learning9
mle9
Programming languages (3)
Github contributions (5)
pytorch/opacus
Feb 2020 - Jan 2023
Training PyTorch models with differential privacy
Role in this project:
Back-end Developer & ML Engineer
Back-end Developer & ML EngineerContributions:10 releases, 202 reviews, 109 commits in 2 years 11 months
Contributions summary:Karthik focused on enhancing the `pytorch/opacus` repository, a project for training PyTorch models with differential privacy. Their contributions included implementing and supporting new functionalities for `Conv1d` layers and enhancing the existing code for convolutional layers to handle non-default stride and padding. They also addressed the handling of frozen layers to improve efficiency and contributed to the core functionalities of computing per-sample gradients and the related test files.
pytorchpytorch-modelsprivacydeep-learningprivacy-preserving-machine-learning
facebookresearch/pytext
Nov 2019 - Nov 2021
A natural language modeling framework based on PyTorch
Role in this project:ML Engineer
ML EngineerContributions:5 commits, 3 PRs in 2 years
Contributions summary:Karthik implemented and extended metrics related to precision and recall for the PyText framework, including the `precision_at_recall` metric and integration into the `SoftClassificationMetrics` structure. They added API support to include the `privacy_engine` in the `report_metric()` function within the metric reporters. Also included is the addition of differential privacy to the PyText framework, through integration with `torchdp` library. Additionally, the user refactored the data sharder configuration.
pytorchnlpbertmachine-learningnatural-language
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.