Experience

Work Experience

Applied Scientist, Amazon Sustainability August 2024-Present

• Designed a Large Language Model (LLM)-based mapping co-pilot to automate manual audit annotation process, leading to 51% savings in processing time.
• Led the development of an AI-based risk assessment tool for Amazon supplier audit optimization, leading to 20% reduction of audits at low-risk sites, with ~$3.5M in savings.

Quantitative Researcher, Cubist Systematic Strategies August 2023-August 2024

• Led a team of 3 researchers on using Language Models for systematic trading signal generation achieving a Sharpe 4 signal using textual data from broker reports, earning reports, and conference call transcripts.
• Designed a fast and distributed ML model fitting framework for financial signal and observed returns regression, leading to an average 35% improvement over existing baseline. Implemented the above framework to run at scale, optimizing the compute and memory management.
• Developed a backtesting platform for trading research, incorporating supervised and unsupervised metrics.Conducted an extensive survey on a new area of research- Foundation Models (FM) and its implementation.

Intern- Product RD and Infrastructures- US Applied Research Center, ByteDance, Mountain View, CA, May–August 2022

• Conducted an extensive survey on a new area of research- Foundation Models (FM) and its implementation.
• Proposed a research approach to using FMs in a federated learning setting, combining distributed training, aggregation, and online temporal updating of FMs.
• Designed a proof-of-concept implementation of above approach to introduce high performance and communication-efficient federated multi-task learning for ByteDance products.

• Worked on the NLP engine of Alexa to enhance its contextual question-answer capability.
• Implemented Contextual Bandits and Experience Replay to improve follow-up question-answering by 17%.
• Introduced and implemented Online Learning pipeline for continual training and adaptation of the model.

• Developed sink interface of Embedded Display Port (eDP) protocol for energy-efficient single stream transport using System Verilog and C++.
• Improved the simulation verification mechanism by developing a universal check script in Perl.

Teaching Experience

• Head of Content Development, Course on Optimization Models in Engineering (EECS 127/227AT), UC Berkeley, with Prof. Venkat Anantharam, Fall 2020


• Content Developer, Course on Optimization Models in Engineering (EECS 127/227AT), UC Berkeley, with Prof. Alexandre M. Bayen, Fall 2019

Professional Services

• Lead Organizer: Tackling Climate Change with Machine Learning at ICML 2021.


• Organizer: Tackling Climate Change with Machine Learning at NeurIPS 2021.


• Publicity Chair: International Conference on Systems for Energy-Efficient Built Environments (BuildSys) 2022.


• Lead Organizer: International Workshop on Applied Machine Learning for Intelligent Energy Systems at ACM e-Energy Conference 2019, 2020, 2021, 2022, 2023.


• Organizer: 1st Climate Change AI Summer School on Tackling Climate Change with Machine Learning, 2022.


• Reviewer: AAAI Conference on Artificial Intelligence 2022, 2023.


• Reviewer: Tackling Climate Change with Machine Learning workshops at NeurIPS 2019, ICLR 2020, NeurIPS 2020.


• Meta Reviewer: Tackling Climate Change with Machine Learning workshops at ICML 2021, NeurIPS 2021.


• Reviewer: Multimodal Learning and Applications workshop at CVPR 2021, 2022


• Reviewer: European AI4Cities Grant Program


• Reviewer: Journal of Building Engineering