Working with IESL lab at UMass Amherst

Sep 2023 - Dec 2023

• Finetuned Language Models to improve zero-shot decision-making performance of RL agents by 33%

• Performed error & qualitative analysis on model to show the validity of our approach.

Work got accepted at NAACL 2024 Findings Conference

Jan 2024 - Present

• Investigating utility of latent information of future tokens in improving the language generation beyond the autoregressive inference framework

• Orchestrated optimization of Yahoo Map Search’s autoscaling services, leveraging time series forecast predictions to increase resource utilization by 20% and significantly reduce operational expenditure by 96%

• Built data cleaning procedures and slack scrapers to gather incident data for training Summarization LLM’s

• Designed and implemented a microservice-based backend for user-website and doctor android apps of Snazzy.

• Developed a general backend for admin-portal for internal management of all the users and deployed it successfully across the organisation. Currently the integrated backend serves about 50+ users and works with a latency of about 50ms in average.

• Worked with Golang, AWS and wrote the backend systems for remote patient monitoring from scratch.

• Led the backend team of about 5 engineers and decided on the optimal tech stack, database schema design and the deployment pipeline.

Under the guidance of Dr. Srijith P.K. and Dr. Maunendra Desarkar, worked on developing semi-supervised models to apply knowledge distillation of privileged features to predict using only commonly available features of community question answer datasets.

This work has been selected as a short paper at the KDIR 2022 conference.