Biography
I'm an engineering manager at AWS in Seattle, where I work on Amazon Quick. My team builds the desktop app, which brings its AI agents to your own computer. Before that we built Quick Research, the deep-research agent.
I've been at Amazon since 2016. I started on supply chain optimization, then spent seven years on Just Walk Out, the technology behind checkout-free stores. There I went from running computer vision models at scale, in the cloud and on in-store devices, to leading the teams that built its machine learning platform. Along the way I co-wrote ORL, a paper on benchmarking reinforcement learning on problems like bin packing and vehicle routing.
At home I'm teaching skills to my robotic arm, and writing about it as I go. I have a master's in computer science from NYU.
Interests
- Robot learning
- Agentic AI
- Computer vision
- Reinforcement learning
Education
- M.S. Computer Science, 2015
New York University · GPA 3.97/4.0 · fellowship - Bachelor's, Information Technology, 2013
Manipal Institute of Technology · GPA 8.23/10
Projects
all projectsWritings
all writingsThe first posts are on their way, starting with the Roll Model series.
News
all news-
Oct 2026
orange-bot picked the right die in 19 of 20 formal trials at night (95%; Wilson interval 76–99%).
-
Oct 2026
Promoted to Senior Software Engineering Manager at AWS.
-
Sep 2026
The Quick desktop app is now generally available on Mac and Windows.
-
Apr 2026
We launched the Quick desktop app in preview, on Mac and Windows.
Publications
all publications-
2019
ORL: Reinforcement Learning Benchmarks for Online Stochastic Optimization Problems
Reinforcement Learning (RL) has achieved state-of-the-art results in domains such as robotics and games. We build on this previous work by applying RL algorithms to a selection of canonical online stochastic optimization problems with a range of practical applications: Bin Packing, Newsvendor, and Vehicle Routing. While there is a nascent literature that applies RL to these problems, there are no commonly accepted benchmarks which can be used to compare proposed approaches rigorously in terms of performance, scale, or generalizability. This paper aims to fill that gap. For each problem we apply both standard approaches as well as newer RL algorithms and analyze results. In each case, the performance of the trained RL policy is competitive with or superior to the corresponding baselines, while not requiring much in the way of domain knowledge. This highlights the potential of RL in real-world dynamic resource allocation problems.
@article{balaji2019orl, title = {{ORL}: Reinforcement Learning Benchmarks for Online Stochastic Optimization Problems}, author = {Balaji, Bharathan and Bell-Masterson, Jordan and Bilgin, Enes and Damianou, Andreas and Garcia, Pablo Moreno and Jain, Arpit and Luo, Runfei and Maggiar, Alvaro and Narayanaswamy, Balakrishnan and Ye, Chun}, journal = {arXiv preprint arXiv:1911.10641}, year = {2019} } -
2013
Most Influential Contextual-Features (MICF) based model for Context-Aware Recommender System
@inproceedings{rana2013micf, title = {Most Influential Contextual-Features ({MICF}) based model for Context-Aware Recommender System}, author = {Rana, Satakshi and Jain, Arpit and Panchal, V. K.}, booktitle = {2013 International Conference on Emerging Trends in Communication, Control, Signal Processing and Computing Applications (C2SPCA)}, pages = {1--6}, year = {2013}, publisher = {IEEE}, doi = {10.1109/C2SPCA.2013.6749418} } -
2013
Contextual information based recommender system using Singular Value Decomposition
@inproceedings{gupta2013contextual, title = {Contextual information based recommender system using Singular Value Decomposition}, author = {Gupta, Rahul and Jain, Arpit and Rana, Satakshi and Singh, Sanjay}, booktitle = {2013 International Conference on Advances in Computing, Communications and Informatics (ICACCI)}, pages = {2084--2089}, year = {2013}, publisher = {IEEE}, doi = {10.1109/ICACCI.2013.6637502} } -
2013
A novel method to represent repetitive and overwriting activities in motion history images
@inproceedings{gupta2013mhi, title = {A novel method to represent repetitive and overwriting activities in motion history images}, author = {Gupta, Rahul and Jain, Arpit and Rana, Satakshi}, booktitle = {2013 International Conference on Communication and Signal Processing (ICCSP)}, pages = {556--560}, year = {2013}, publisher = {IEEE}, doi = {10.1109/ICCSP.2013.6577116} }
Experience
-
2021 – now
Amazon Web Services Seattle-
Senior Software Engineering ManagerOct 2026 –
Amazon Quick as a desktop app for Mac and Windows, so its AI agents can work alongside you on your own computer. Preview in April 2026, generally available in September.
-
Software Development ManagerMar 2025 – Oct 2026
Quick's deep-research agent. Ask it a question and it comes back with a full report, sources included. Launched with Quick Suite in October 2025.
-
Software Development Manager2021 – 2025
Led the machine learning engineering and MLOps teams behind its computer vision models. We set the training and data strategy, then built and scaled the organization's ML platform, and I co-wrote its two-year MLOps vision.
-
-
2015 – 2021
Amazon Seattle-
Software Development Engineer II, then Senior SDE2018 – 2021Just Walk Out
Large-scale inference hosting for computer vision models, in the cloud and on edge devices in the stores.
-
Software Development Engineer, then SDE II2016 – 2018Supply Chain Optimization Technologies
Linear programming for supply-chain optimisation.
-
Software Development Engineer InternSummer 2015Supply chain management
-
-
2013
Defence Research and Development Organisation Delhi-
Research InternJun – Dec 2013
Combined deep learning with swarm intelligence to classify land cover in satellite images.
-