// hello world

I'm Ashish Kumar
Uchadiya

AI & Machine Learning Researcher

M.Tech — Industrial Engineering & Operations Research, IIT Bombay

Researching efficient multi-modal generative models, computer vision, and deep learning. Building intelligent systems that bridge cutting-edge research and real-world impact.

Who I Am

I'm a Master's student at IIT Bombay in Industrial Engineering & Operations Research, working on my thesis in Efficient Multi-Modal Generative Models under the guidance of Prof. P Balamurugan.

My research spans GANs, Diffusion Models, computer vision, and medical image analysis. I've achieved 1st rank in the AI/ML Medical Image Championship at IIT Bombay and worked on large-scale challenges including a NeurIPS 2024 Kaggle competition processing 300M+ molecular samples.

I also serve as a Teaching Assistant for Machine Learning and Deep Learning courses, and enjoy Python automation, video encoding, and GPU-accelerated computing.

Master's Thesis

Efficient Multi-Modal Generative Models — GANs, Diffusion Models, high-res image generation

Teaching Assistant

IE506 Machine Learning & IE643 Deep Learning at IIT Bombay

Competitions

1st Rank — AI/ML Medical Image Championship, IIT Bombay | NeurIPS 2024 Kaggle | Amazon ML Challenge 2024

Academic Background

Master of Technology

Industrial Engineering & Operations Research

Indian Institute of Technology, Bombay

2023 — 2025

CPI: 8.22

Bachelor of Engineering

Electrical Engineering

SATI Vidisha — RGPV Bhopal

2018 — 2022

CPI: 7.83

Teaching Assistant — IE506 Machine Learning Ongoing

Instructor: Prof. P Balamurugan · IIT Bombay

Teaching Assistant — IE643 Deep Learning

Instructor: Prof. P Balamurugan · IIT Bombay · Spring 2024

What I Work With

ML / Deep Learning

PyTorch Scikit-Learn OpenCV Optuna PySpark

Languages & Data

Python SQL MATLAB Pandas NumPy

Tools & Platforms

Linux Git PowerBI Excel Figma

Key Courses

Machine Learning Deep Learning Math Optimization Probability & Stochastics Engineering Statistics

What I've Built

Research projects, competition entries, and self-driven work.

Jul 2024 — Ongoing

Efficient Multi-Modal Generative Models

Master Thesis · Prof. P Balamurugan

Experimenting with GANs (DCGAN, Conditional-GAN) and exploring Diffusion Models to enhance high-resolution image generation while reducing training time and hardware requirements.

PyTorch GANs Diffusion CUDA

Feb 2024 — Mar 2024

AI/ML Medical Image Championship

KCDH & Web Club, IIT Bombay

Achieved 1st rank in skin disease classification. Used EfficientNet with fold training, class weights, and ensembling to reach 79.35% balanced accuracy.

EfficientNet PyTorch Ensembling

Dec 2024 — Jan 2025

Protein Binding Affinity Prediction

NeurIPS 2024 · Kaggle Challenge

Processed 300M+ molecular samples in SMILES representation. Calculated molecular fingerprints and applied XGBoost & MLP, achieving 0.24 mAP score across three protein targets.

PySpark XGBoost MLP SMILES

Jan 2024 — Apr 2024

Tumor Localization & Segmentation

M.Tech Seminar · Prof. P Balamurugan

Implemented a Modified UNet with skip connections for brain MRI tumor mask prediction. Achieved a Dice score of 74.4% by epoch 10 with optimized computation.

UNet PyTorch Medical Imaging

Oct 2024

Product Entity Extraction — Zero-Shot

Amazon ML Challenge 2024

Extracted product entities from 6000+ images using pretrained OCR and zero-shot prompting, handling diverse image orientations for automated entity extraction.

OCR Zero-Shot Python

Jul 2024 — Aug 2024

Custom Chatbot with Fine-Tuned LLMs

Self Project

Fine-tuned GPT-2 & LLaMA-3.1 on domain-specific books. Built a training pipeline handling PDF, DOCX, and TXT formats for domain-adapted conversational AI.

LLaMA-3.1 GPT-2 Fine-Tuning LLMs

Jan 2024 — Apr 2024

Outlier Detection & Robust PCA

Course Project · Machine Learning

Formulated feature-level outlier detection using Robust PCA as a convex optimization problem. Tested on synthetic datasets and YouTube videos to identify outlier frames.

Robust PCA Convex Optimization Python
View All on GitHub

Let's Connect

Open to research collaborations, open-source contributions, and conversations about AI/ML.