Amazon Artificial General Intelligence (AGI) - Customizations

Senior Applied Scientist
Bellevue, WA
February 21 - Present
* Led the development and launch of Amazon Nova customizations features, implementing advanced fine-tuning recipes including DPO (Direct Preference Optimization), PPO (Proximal Policy Optimization), and GRPO (Group Relative Policy Optimization). * Designed and optimized reinforcement learning from human feedback (RLHF) pipelines to enhance model alignment and performance for enterprise customer use cases. * Collaborated with cross-functional teams to integrate customization capabilities into Amazon Nova, enabling customers to fine-tune foundation models for their specific domains and requirements. * Contributed to the technical architecture and scalable deployment of fine-tuning infrastructure supporting multiple optimization techniques across AWS's AI services.

2. Amazon.com

Machine Learning Engineer Intern
MA/ Virtual
May 2020 - September 2020
Worked on personalized ranking for Amazon.com customers, providing personalized content ranking displayed on Amazon’s homepage for customers based on various individual metrics.

3. Samsung Semiconductor India Research

Machine Learning Engineer
Bangalore, India
July 2017 - December 2018
Developed an LTE Physical Layer simulator and performed cache optimization of LTE DSP code, resulting in a successful reduction of memory utilization and improvement in cache hit rate.

Research Experience

1. Information Extraction & Synthesis Laboratory (I.E.S.L)

UMass Amherst
September 2020 - February 2021
Implemented unsupervised approaches to meta-learning in order to improve few-shot generalization of NLP. Evaluated multiple pre-trained models for few-shot generalization to new tasks and new domains.

2. Information Extraction & Synthesis Laboratory (I.E.S.L)

UMass Amherst & Chan Zuckerberg Initiative
January 2020 - May 2020
Developed a BERT-based architecture and used BioSentVec embedding for Biomedical Question Answering with CZI. The model saves significant amounts of time for medical practitioners searching patient records and drug interactions.

3. Veterinary & Animal Sciences Department (VASCI)

UMass Amherst
May 2019 - December 2019
Used machine learning for prediction and annotation of DNA sequences crucial for genes. The research project was funded by the USDA National Institute of Food and Agriculture (NIFA).

4. The Laboratory for Mobile Sensing and Ubiquitous Computing (MOSAIC),

UMass Amherst
May 2019 - August 2019
Developed a novel model with Prof. Tauhidur based on multispectral physiological parameter estimation. Researched the effect of external factors such as angle, blur, lighting, and sweat on the estimation project. Worked on a novel LSTM architecture for the estimation of Galvanic skin response with multispectral imaging.