Feb 2020 – Dec 2024 · 4 yrs 10 mos · San Francisco Bay Area

• I was one of the initial members of Store Foundational AI (SFAI-M5) where we trained domain specific model to power multiple vertical applications like search , advertising and catalog management. During these years I worked on pre-training, post-training, distillation, efficient distributed training and inference optimisation to enable LLMs on Web-scale applications. My projects led to $XXX M annual search-attributed sales and $XX M annual revenue boost in Advertising.

May 2019 – Aug 2019 · 3 mos · San Francisco Bay Area

• Interned at Amazon Shopping & Discovery team [in Palo Alto], to improve semantic product matching by incorporating product-query purchase graph in deep learning models.

Jan 2019 – May 2019 · 4 mos

• Teaching Assistant for the course 11-791 - Design of Intelligent Information Systems (Spring 2019) taught by Professor Eric Nyberg. The aim of the course is to extrapolate the fundamentals of system engineering (requirements, analysis, design, implementation) and project management (teaming, planning, scheduling, tracking) to Machine learning and Artificial Intelligence pipelines. My role was to design and grade the assignments for the course.

Sep 2018 – May 2019 · 8 mos

• Working with Prof. Eric Nyberg on Neural Question Answering(QA) Systems and QA agent architectures.
• Investigating "A Question-Focused Multi-Factor Attention Network for Question Answering"(AMANDA) neural QA model and adapting it to different datasets, to see it's performance. The findings across different models will later be used to summarize their generalization ability.
• Worked on Scraping and Entity resolution component of Text Graph, which is used to identify good passage candidates for a specific inference type for Question Generation.

Aug 2016 – Jul 2018 · 1 yr 11 mos · Bangalore

• Prototyped an unsupervised learning system on mixed-attributes using K-modes and K-prototypes clustering algorithm for analyzing trade breaks
• and fails, resulting in identification of 5 issues which led to 40% decrease in analyzed break types.
• Led development of an Information Retrieval system for Firmwide Business Intelligence(BI) Portal, enabling fine-grained search on dashboard components to locate specific business metrics efficiently. Worked along a Summer Intern and a team of under-graduate hires (as a part of training project) to successfully complete the project.

May 2015 – Jul 2015 · 2 mos · Bangalore

May 2014 – Jul 2014 · 2 mos

• I added features to BureauBuilder - digital solutions for matchmakers and marriage bureaus.The project was done on Ruby On Rails web-framework .