Research & Publications

Mix of research work, publications, and applied internship projects. Full details are in my resume!

EEG-based detection of secondary brain injury

May 2021 – Jun 2022

Center for the Neural Basis of Cognition, Carnegie Mellon University

Graduate Research Assistant, Deep Learning Researcher

At CMU, I worked on EEG-based, non-invasive detection of secondary brain injury in traumatic brain injury and stroke. The project focused on identifying cortical spreading depression from EEG recordings using a mix of graph signal processing ideas and deep learning. I developed and iterated on deep learning models including ResNet34 and Graph Attention Networks, and I scaled experiments through GPU training and high-performance computing to run larger, more reliable evaluations.

Financial data modeling for equity premium prediction

Jan 2022 – May 2022

Carnegie Mellon University

Data Science Intern, Financial Data Modeling

At CMU, I worked on financial signal engineering and rolling predictive modeling for equity premium prediction. I engineered signals such as dividend yield and term spread, improving data reliability and reducing missingness by 25%. I built and tuned rolling prediction pipelines that improved R² from 0.55 to 0.78 and reduced MSE by 40% through feature selection and hyperparameter tuning. I also used bootstrapping and signal extraction techniques to stabilize forecasts under noisy market conditions.

Protein subcellular localization from microscopy images

Jan 2019 – Jun 2019

Intelligent Systems Center, PES University

Machine Learning Researcher

This project focused on automating protein subcellular localization from microscopy images to reduce reliance on slow, manual annotation. I framed the task as a multi-class image classification problem and ran a structured comparative study across three model families: a conventional baseline, a ResNet50-based architecture, and an Xception-based architecture. By evaluating these approaches side by side, I quantified the performance gains from modern deep learning architectures and identified the most reliable model for scaling the localization pipeline. The work received the Best Research Project and Innovation award, and I first-authored the resulting paper.