Sep 2024 – Dec 2024 · 3 mos

Jun 2024 – Aug 2024 · 2 mos

• Title: Stance Detection from Sustainability News and Reports
• Engineered a custom Retrieval-Augmented Generation (RAG) pipeline using LLMs for policy document analysis, extracting key information such as focus areas, budgets, and subtasks.
• Developed advanced NLP solutions for news article processing, including classification, entity extraction (key actors, beneficiaries, and financial data), and automated summarization.
• Implemented a web scraper to efficiently collect topic-specific articles from Indian sources over defined time periods, enhancing the NLP pipeline with comprehensive and timely data.

May 2024 – Dec 2024 · 7 mos · Remote

• Selected as Volunteer as a Research Assistant at the Financial Services Innovation Lab, Georgia Institute of Technology.

Jan 2024 – Mar 2024 · 2 mos · On-site

• Title: Oral Cancer Detection using Computer Vision and Histopathology Images
• Optimized CellVit, a cutting-edge computer vision model, to enable patch-level, zero-shot inference, enhancing the accurate detection of oral cancer cells from histopathology images.
• Designed the data processing pipeline, including image reshaping and stain normalization, to ensure seamless integration of diverse histopathology datasets into the CellVit pipeline.
• Developed and validated classification models to differentiate between three stages of cancer—Normal, Oral Submucous Fibrosis (OSMF), and Oral Squamous Cell Carcinoma (OSCC)—achieving 85% accuracy on the validation set, in collaboration with pathologists to cross-verify results.
• Implemented an automated pipeline to extract patches from 100x high-resolution histopathology images, apply stain normalization, and integrate them into classification models for training and testing.

Jul 2023 – Sep 2023 · 2 mos · Hybrid

• Working on developing deep learning algorithms to predict gene expression levels in Yeast and T-cells from DNA promoter sequence under Mphasis Lab (ML2CT)
• Mentor: Professor Rintu Kutum (Ashoka University)
• Funding Agency: Mphasis

Jul 2023 – Aug 2023 · 1 mo · Hybrid

• Leveraged state-of-the-art transformer-based semantic segmentation models for accurate land usage prediction from satellite images, contributing to improved categorization accuracy.
• Developed code to extract smaller patches from large satellite images, enhancing processing efficiency for the semantic segmentation model.
• Created custom code to convert multi-channel RGB masks into single-channel semantics masks for cross-entropy loss calculations, and vice versa, enabling seamless integration of the model with the provided data format.