Praj B
AI Resident at Meta
New York, New York, United States
Summary
🤩
Rockstar🎓
Top SchoolPraj B is a research engineer with nine years of experience specializing in pre-training and scaling large language models, currently contributing to Llama 4 long-context pretraining and inference optimizations at Meta. They bring deep hands-on expertise across conversational AI, decision-making models, and generative LLMs from roles as an AI Resident and core contributor on projects like ParlAI and Hugging Face Transformers. Praj has a track record of production-focused improvements—ranging from DST features in dialogue systems to native AMP and training refinements in transformers—and has helped demonstrate model capabilities through PyTorch tutorials. Based in New York with an MS thesis in computer science, they blend research rigor with practical infra and evaluation work, notably enabling 10M+ token long-context modeling and tooling that eases adoption by the broader ML community.
9 years of coding experience
Master of Science - MS (Thesis), Computer Science, Master of Science - MS (Thesis), Computer Science at The University of Texas at DallasDeep Learning Specialization, Deep Learning, 96.5, Deep Learning Specialization, Deep Learning, 96.5 at CourseraFastai Deep Learning (Part 2), International Fellow, Fastai Deep Learning (Part 2), International Fellow at University of San Francisco
English, German
Github Skills (15)
transformers10
computer-vision10
pytorch10
machine-learning10
nlp10
deep-learning10
trainings10
python10
natural-language-processing10
modeling10
ml9
aiml9
ai9
code-optimization9
tensorflow5
Programming languages (9)
Github contributions (5)
facebookresearch/ParlAI
Jun 2022 - Aug 2022
A framework for training and evaluating AI models on a variety of openly available dialogue datasets.
Role in this project:
ML Engineer
ML EngineerContributions:4 reviews, 8 commits, 14 PRs in 2 months
Contributions summary:Praj primarily contributed to the `parlai` repository, a framework for dialogue AI. Their work involved implementing and refining dialogue state tracking (DST) features for the Multiwoz_v22 task. This included creating a DST teacher, incorporating evaluation metrics, and fixing seed-related issues, as well as addressing minor documentation typos. The user also exposed certain functionalities to the user, such as precision and recall in teacher metrics and the option to retain a graph, for a Torch agent.
nlpdeep-learningdatasetmachine-learningtraining
huggingface/transformers
May 2020 - Sep 2020
🤗 Transformers: State-of-the-art Machine Learning for Pytorch, TensorFlow, and JAX.
Role in this project:ML Engineer
ML EngineerContributions:5 commits, 11 PRs, 28 comments in 3 months
Contributions summary:Praj made several contributions focused on improving the `transformers` library, specifically related to Pytorch and deep learning. Their work included refactoring and updating code to utilize the latest Pytorch features such as native AMP support. They also fixed typos and removed redundant arguments in key training and modeling files, demonstrating a focus on code quality and optimization within the machine learning framework.
pythonbertspeech-recognitionstate-of-the-artflax
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.