Dustin Tran
Member Of Technical Staff at xAI
Mountain View, California, United States
Summary
🤩
Rockstar🎓
Top SchoolDustin Tran is a research scientist and ML engineer with 13 years of experience building foundational models, probabilistic systems, and evaluation frameworks; he recently joined xAI to lead the post-training team after a senior research role at Google DeepMind. He co-created Gemini-Exp-0801 and is a core contributor to Gemini 1 through 2.5, driving work across RL, data, and evaluations that produced top-ranked models on LMSYS, WebDev Arena, and HLE. His open-source footprint spans TensorFlow projects (tensor2tensor, Mesh TF, TensorFlow Probability), Edward/Edward2, Stan, and Google Research—demonstrating deep expertise in probabilistic programming, variational inference, and scalable model tooling. Trained at Columbia, Harvard, and UC Berkeley, he combines rigorous academic grounding with hands-on production engineering, and uniquely pairs contributions to core loss functions and CI/build automation with leadership of large-scale model evals.
13 years of coding experience
7 years of employment as a software developer
Bachelor of Arts (B.A.) Mathematics Statistics, Bachelor of Arts (B.A.) Mathematics Statistics at University of California, BerkeleyDoctor of Philosophy (Ph.D.) Statistics, Doctor of Philosophy (Ph.D.) Statistics at Harvard UniversityDoctor of Philosophy (Ph.D.) Computer Science, Doctor of Philosophy (Ph.D.) Computer Science at Columbia University
Stackoverflow
479reputation
19kreached
2answers
6questions
Github Skills (38)
bayesian-statistics10
c-language10
probabilistic-programming10
variational-inference10
python10
bayesian-data-analysis10
data-science10
machine-learning10
bash10
cicd10
autoencoder10
bayesian10
deep-learning10
tensorflow10
neural-network10
Programming languages (7)
Github contributions (5)
blei-lab/edward
Feb 2016 - Dec 2018
A probabilistic programming language in TensorFlow. Deep generative models, variational inference.
Role in this project:
ML Engineer
ML EngineerContributions:29 releases, 1640 commits, 539 PRs in 2 years 10 months
Contributions summary:Dustin's commits center around the development of new model structures. The user implemented a metagraph implementation, including support for random variables and distributions, to be utilized in the probabilistic programming language Edward. The updates demonstrate an understanding of TensorFlow's stochastic graph capabilities to build and train new model structures using the framework. The contributions introduce support for diverse distribution families to Edward's functionality.
inferenceneural-networksmachine-learningprobabilistic-programmingdeep-generative-models
google/uncertainty-baselines
Aug 2020 - Jul 2022
High-quality implementations of standard and SOTA methods on a variety of tasks.
Role in this project:ML Engineer & Data Scientist
ML Engineer & Data ScientistContributions:50 reviews, 98 commits, 6 PRs in 1 year 11 months
Contributions summary:Dustin primarily focused on implementing and adapting machine learning models for image classification tasks, particularly within the context of the Uncertainty Baselines project. The user's contributions included porting Edward2 baselines, integrating new uncertainty metrics from github.com/google/uncertainty-metrics, and modifying the training pipeline to accommodate new models and datasets. Furthermore, the user addressed code organization by restructuring modules for different uncertainty metrics.
implementationsstatisticsdata-sciencedeep-learningneural-networks
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.