Apr 2021 – Nov 2025 · 4 yrs 7 mos · Bengaluru, Karnataka, India · On-site

• Inference latency optimization of transformer models in the Search team:
• Used inference latency profiles of transformer models derived from roofline analyses to identify the most promising combination of model compression and inference optimization techniques for experimentation and implementation on target inference hardware.
• Reduced the inference latency of encoder-only and encoder-decoder transformer models used in search query annotation and reformulation with a combination of weight pruning and post-training quantization.
• Reduced the end-to-end text generation latency of a large language model (LLM) used in anaphoric and contextual dereferencing of customer questions in a multi-turn setting with a combination of prompt caching and compression, post-training quantization, and speculative decoding that reduced the LLM's time to first token as well as subsequent inter-token latency.
• A part of this work was published at the Amazon Machine Learning Conference (AMLC) 2024.
• Developed machine learning models for various applications in the Search team:
• Multi-objective search relevance ranking with transformer based ranking models.
• Low latency search query annotation with LLM based classification models.
• Identifying product attributes along which to organize search results into themed containers in order to improve customer search experience while navigating search results.
• Identifying vertical search intent in queries to improve customer search experience with transformer based classification models.
• Expanding coverage of numeric behavioral features used in search relevance ranking with transformer based regression models.
• Machine learnt generic keywords to improve discoverability of new products in search.

Jul 2015 – Apr 2021 · 5 yrs 9 mos · Bengaluru, Karnataka, India · On-site

• Developed machine learning models for various applications in the Search team:
• Product popularity scores for ranking of products in search and browse results.
• User cohort based personalization of product ranking in search and browse results.
• Catalog attribute identification for search query tokens to improve matching and retrieval of relevant products. This work was published as a full paper at ECIR-2020.
• Click models for quantifying presentation bias in the click logs and evaluation of product ranking functions via offline simulations.
• Catalog category identification for search queries to improve retrieval quality and reduce retrieval latency.
• Spelling correction for search queries.
• Developed machine learning models for various applications in the User Insights team:
• Customer lifetime value modelling and prediction.
• Lookalike audience creation for targeted marketing and advertising.
• Embedding vector representation of users for various insight prediction tasks.

Jun 2012 – Sep 2012 · 3 mos · Austin, Texas, United States

• Developed classification algorithms for detection of fraudulent transactions in Apple Online Store's order processing system.

May 2009 – Aug 2009 · 3 mos · Austin, Texas, United States

• Developed machine learning algorithms for large scale link analysis, text classification, collaborative filtering, and content ranking.

Aug 2005 – Dec 2012 · 7 yrs 4 mos · Austin, Texas, United States

• EE380L Data Mining: Data analysis, clustering, classification, and regression techniques.
• EE360C Algorithms: Sorting and searching algorithms, graph algorithms, algorithm design techniques, and complexity theory.
• EE319K Introduction to Microcontrollers: Conducted weekly labs emphasizing core concepts in embedded systems, assembly programming, and interfacing I/O devices.