Tolga Bolukbasi
Member Of Technical Staff at Reflection AI
San Francisco, California, United States
Summary
🤩
Rockstar🎓
Top SchoolTolga Bolukbasi is an ML researcher-engineer with 12 years of experience specializing in large-scale pretraining, model interpretability, and scalable LLM systems, currently a founding member of the pretraining team at Reflection AI. He was a core contributor to Gemini 1.0–3.0 pretraining at Google/DeepMind, driving work on distillation, deduplication, value-of-data, and reliable evaluation frameworks that remained private until release. His background spans research on data attribution and gradient-based explainability, practical saliency and interpretability tooling, and compute-efficient inference techniques like speculative decoding and cascaded networks. A hands-on open-source contributor, he has improved widely used tools such as the What-If Tool, LIT, and TensorBoard integrations to make model analysis and fairness diagnostics more accessible. He holds a PhD in Electrical Engineering from Boston University and combines deep theoretical insight with production-grade engineering across both research and developer tooling. An under-the-radar strength is his track record of turning interpretability research into usable demos and integrations that bridge research and product needs.
12 years of coding experience
8 years of employment as a software developer
ERASMUS Exchange, ERASMUS Exchange at Tampere University of Technology 1965-2018High School, High School at Istanbul LisesiBachelor of Science (BS), Electrical and Electronics Engineering, Bachelor of Science (BS), Electrical and Electronics Engineering at Orta Doğu Teknik Üniversitesi / Middle East Technical UniversityDoctor of Philosophy (Ph.D.), Electrical Engineering, Doctor of Philosophy (Ph.D.), Electrical Engineering at Boston University
Turkish, English, German
Github Skills (28)
javascript10
visualization10
python10
machine-learning10
typescript10
keras10
deep-learning10
tensorflow10
deep-neural-networks10
typescript-types10
neural-network10
typescripts10
visualizations10
react10
apidoc9
Programming languages (6)
Github contributions (5)
PAIR-code/saliency
Jul 2019 - Oct 2020
Framework-agnostic implementation for state-of-the-art saliency methods (XRAI, BlurIG, SmoothGrad, and more).
Role in this project:
ML Engineer
ML EngineerContributions:18 reviews, 9 commits, 21 PRs in 1 year 2 months
Contributions summary:Tolga primarily contributed to the implementation and improvement of saliency methods for computer vision. Their work included reverting and fixing bugs related to integrated gradients and SmoothGrad, along with adding the XRAI saliency algorithm. They also updated the project's dependencies and fixed import issues related to TensorFlow 2.x. Further, they bumped the project's version and added a new feature.
pytorchartimage-recognitionsaliency-mapdeep-learning
PAIR-code/what-if-tool
Feb 2019 - Sep 2020
Source code/webpage/demos for the What-If Tool
Role in this project:ML Engineer
ML EngineerContributions:20 reviews, 17 commits, 8 PRs in 1 year 7 months
Contributions summary:Tolga primarily contributed to the What-If Tool, focusing on the integration and demonstration of machine learning models. Their work involved updating and refactoring existing notebooks (e.g., `WIT_Toxicity_Text_Model_Comparison.ipynb`) to use a new prediction function API, and adding new demo notebooks (e.g., `WIT_Smile_Detector.ipynb`) showcasing image-based model analysis. They implemented custom prediction functions for the Keras models, demonstrating an understanding of model integration within the What-If Tool framework.
webpagejupyterlab-extensionwhat-ifmachine-learningtensorboard
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.