Hi, I'm Adam Kraft.
Research Engineer at Google DeepMind.
Professional Experience
Research Engineer
Google Brain / Google DeepMind• 2019 - Present
Working on research and applied machine learning / AI with several Google product areas (Gemini, Google Ads, Google Play, Google Search, and Google Cloud). Specializing in Personalization for Large Language Models (LLMs) and Large Recommendation Models (LRMs), Data Efficiency, and Model Efficiency.
Machine Learning Engineer
Orbital Insight • 2015 - 2019
Used large scale Machine Learning to analyze millions of satellite and remote sensing images. Applications included counting cars in parking lots to measure economic activity, tracking new development of buildings and roads, monitoring oil tank levels, and tracking natural resources (water, forests).
Work Topics: Computer Vision, Machine Learning and Deep Learning, Segmentation (Semantic and Instance), Image Classification, Object Detection, Domain Adaptation and Transfer Learning
Machine Learning Software Engineer
A9 / Amazon • 2012-2015
Part of the Visual Search research team in Palo Alto, helping customers search for millions of Amazon products through the “Search by Camera” feature. I led the “Scan to Claim” Gift Card feature from research and development all the way to production. Other projects also used computer vision and machine learning to help power similarity search and other web shopping applications.
Work Topics: Computer Vision, Machine Learning and Deep Learning, Image Classification, Image Segmentation, Instance Matching, Similarity Matching and Category Recognition, Domain Adaptation and Transfer Learning, Multi-frame tracking and stitching
Selected Papers and Projects
Personal Intelligence securely connects information from apps like Gmail and Google Photos to make Gemini uniquely helpful. It connects Gmail, Photos, YouTube and Search in a single tap to provide better, more personalized answers and proactive suggestions based on your personal context.
Gemini Team, Google Labs
We introduce the Gemini 2.X model family: Gemini 2.5 Pro and Gemini 2.5 Flash, as well as our earlier Gemini 2.0 Flash and Flash-Lite models. Gemini 2.5 Pro is our most capable model yet, achieving SoTA performance on frontier coding and reasoning benchmarks.
Gemini Team, Google
RecSys 2025. We introduce Cross-Batch Aggregate (XBA) loss, a pointwise loss designed to approximate aggregate loss in streaming environments where event-level labels are missing.
Jonathan Valverde, Tiansheng Yao, Xiang Li, Yuan Gao, Yin Zhang, Andrew Evdokimov, Adam Kraft, Samuel Ieong, Jerry Zhang, Ed H. Chi, D. Cheng, Ruoxi Wang
We propose a Simple Training-free Approach for Recommendation (STAR), a framework that utilizes LLMs and can be applied to various recommendation tasks without the need for fine-tuning, while maintaining high quality recommendation performance.
Dong-Ho Lee*, Adam Kraft*, Long Jin, Nikhil Mehta, Taibai Xu, Lichan Hong, Ed H. Chi, Xinyang Yi
Accepted to ACL 2024 (Co-first author). We propose a novel framework that leverages the reasoning capabilities of Large Language Models (LLMs) to enhance personalized recommender systems, improving both recommendation quality and interpretability.
Alicia Y. Tsai*, Adam Kraft*, Long Jin, Chenwei Cai, Anahita Hosseini, Taibai Xu, Zemin Zhang, Lichan Hong, Ed H. Chi, Xinyang Yi
NeurIPS 2020 (Oral Presentation). We introduce PyGlove, a new Python library that reformulates AutoML as an automated process of symbolic manipulation, decoupling the search algorithm, the search space, and the child program.
Daiyi Peng, Xuanyi Dong, Esteban Real, Mingxing Tan, Yifeng Lu, Hanxiao Liu, Gabriel Bender, Adam Kraft, Chen Liang, Quoc V. Le
SIGSPATIAL 2018. We propose a regional aggregation layer, a novel neural network component that allows a pixel-level density estimator to be trained using only coarse-grained density aggregates.
Nathan Jacobs, Adam Kraft, Muhammad Usman Rafique, Ranti Dev Sharma
Talks
Presentations on machine learning, systems, and research.