GK

Exploring the interface of brain, data, and health

Research Experience

Neuromatch Academy · International

Computational Neuroscience Fellow

Period June - July 2025
Dataset Miller Stanford (ECoG)
Domain Neural Signal Processing
Format Remote · International cohort
International Fellowship

Investigating Discriminative Neural Signatures Across Motor Tasks

Miller Stanford ECoG dataset · High-gamma power extraction · Single-trial classification

Central question: Whether discriminative neural signatures could reliably separate distinct motor tasks at the level of individual trials.

Methodology: We analysed neural activity across seven Brodmann areas (BA 2, 3, 4, 6, 9, 43, 45) known to be involved in sensorimotor and prefrontal processing. We extracted high-gamma band power features (70-200 Hz) using Welch's spectral estimation method, then applied PCA to reduce dimensionality before attempting single-trial classification.

Key finding

High-gamma power in Brodmann areas 4 and 6 (primary and premotor cortex) showed the strongest discriminative signal across motor tasks, consistent with their established roles in movement planning and execution. PCA-based reduction preserved sufficient variance for above-chance single-trial classification, suggesting these spectral features carry task-relevant information even without trial averaging.

ECoG Welch's Method PCA Brodmann Areas
View project on GitHub

Research Output

Women's Health Research
Forthcoming · 2026

Phase-based Women's Health Recommendation and Prediction System

Gurjot Kaur, Kushagra Agrawaal, Shweta Agrawaal

Innovations in Healthcare Technology and Management · Bentham Books

Core contribution

A RAG-augmented LLM system that classifies menstrual cycle phases and generates personalised diet and exercise plans tailored to each phase's physiological profile, addressing a gap in healthcare technology where generic advice ignores intra-cycle phase variation.

Developed a FastAPI-based recommendation pipeline that uses large language models with RAG to deliver phase-specific health guidance for women. The system classifies cycle phases from user inputs and retrieves evidence-backed recommendations, improving query resolution efficiency over a baseline generative approach.

My contribution

Designed the RAG pipeline, implemented the LLM integration via the Gemini API, built the FastAPI backend, and conducted the evaluation of recommendation quality across cycle phases.

LLMs RAG Women's Health FastAPI
Yume
Published · 2026

Towards Intelligent Manga Generation: Exploring Diffusion Models, GANs, and Deep Learning Approaches

Sneha, Gurjot Kaur, Madan Lal Saini

2025 2nd International Conference on Recent Trends in Electrical, Electronics and Computing Technologies (ICRTEECT) · Conference Paper

Core contribution

This work proposes and evaluates a webtoon-style manga generation pipeline that integrates Magi layout with Anime Detailer XL LoRA and multi-modal image generation models to automate narrative and artistic synthesis

The paper investigates GANs, CNNs, diffusion models, and multi-modal large language models for AI-driven manga and webtoon creation

My contribution

Model selection and evaluation for manga and webtoon generation, and contributed substantially to the research design, literature review, and experimental analysis in the paper.

GANs Manga generation Diffusion Models

Paper DOI:10.1109/icrteect67512.2025.11448686

What I'm Thinking About

Computational Neuroscience

I'm drawn to the question of how the brain encodes and transforms information — particularly in motor and language-related cortical areas. I want to understand how models of neural activity can be grounded in the biology, not just the statistics.

ML for Healthcare Diagnostics

Medicine generates data that is messy, imbalanced, and high-stakes, exactly the conditions where thoughtful ML design matters most. I'm interested in how we build models that are not just accurate but interpretable enough to be trusted clinically.

Brain-Computer Interfaces

BCIs sit at the convergence of everything I find compelling: neural signal processing, real-time ML, and direct clinical impact for people with motor disabilities.