MS(R) Student @ IIT Bombay

Building AI for
Healthcare & Vision

I am a Master’s student specializing in Health AI research, with a strong foundation in computer vision and deep learning. Previously felicitated as a Research Assistant, I have collaborated closely with experienced researchers on stable architecture generation and vision-based models. My current focus is on the field of domain adaptation , few shot learning and meta learning , particularly in the context of medical imaging . I look forward to colab orating with like-minded individuals and organizations to push the boundaries of AI in Computer Vision and Healthcare.

Souradeep Dutta

Experience

June 2025 – Present

Teaching Assistant (TA)

Indian Institute of Technology, Bombay

May 2024 – Nov 2024

Research & Development Intern

Indian Institute of Technology, Bombay

Focusing on Computer Vision and Remote Sensing applications.

Feb 2024 – April 2025

Research Assistant

IEDC (CEDS), Kolkata

Worked on Computer Vision & Generative AI, collaborating on vision-based models and robust architecture generation.

Jan 2022 – Feb 2023

Junior Researcher

IEM IIC, Kolkata

Focused on Machine Learning & AI fundamentals.

Selected Projects

DeepViolence Detector (Swin3D)

Real-time violence detection using a modified Swin-3D Transformer with GUI support.

PyTorchTransformersVideo AI

Vision-Language Satellite Classification (CoOp)

Applying context optimization (CoOp) to enhance CLIP for satellite image classification.

CLIPPyTorchAerial Vision

Histopathology Image Super-Resolution (SR3)

Diffusion-based SR3 model for super-resolving histopathology images.

Diffusion ModelsSR3Medical AI

Retinal Blood Vessel Segmentation

Deep learning pipeline for vessel segmentation in retinal fundus images.

U-NetMedical Imaging

Reasoning VQA with Knowledge Graphs

Knowledge-augmented VQA system combining KG reasoning with visual understanding.

VQAKnowledge GraphsPyTorch

RxIntelli – Medical NLP Assistant

Semantic search and intelligent medical document interpretation.

NLPFastAPIHealthcare AI

Meta-Learning on Satellite Images

Zero-shot recognition systems for satellite data using meta-learning to enable Open World Learning.

PyTorchMeta-Learning

UAV Weather Forecasting

On ground collected data from UAV. Bi-directional LSTM system for temporal sequencing model. 95.9% accuracy with sub-6ms latency.

LSTMIoT

Publications

2025

Revisiting KRISP: A Lightweight Reproduction and Analysis of Knowledge-Enhanced Vision-Language Models

Author Name: Souradeep Dutta , et al.

BibTex Citation: @misc{dutta2025revisitingkrisplightweightreproduction, title={Revisiting KRISP: A Lightweight Reproduction and Analysis of Knowledge-Enhanced Vision-Language Models}, author={Souradeep Dutta and Keshav Bulia and Neena S Nair}, year={2025}, eprint={2511.20795}, archivePrefix={arXiv}, primaryClass={cs.CV}, url={https://arxiv.org/abs/2511.20795}, }

Blogs

Golden Noise: The Secret Ingredient Making Diffusion Models Smarter

A simple look at ICCV 2025’s breakthrough in learnable noise for diffusion models.

Nov 2025

Technical Skills

Languages

Python, C, C++, R, Java, SQL, MATLAB

Frameworks

PyTorch, TensorFlow , Streamlit

Tools

Google Cloud, AWS, ChromaDB , FAISS, GiT

Awards

  • Top 25 All India Rank - Deep Learning (NPTEL, IIT Madras)
  • Rank 863 Globally - Google Isolated Sign Language Challenge
  • Ranked 5th in IIT Bombay - WnCC x Loop AI/ML Healthcare Hackathon
  • Patent Published - No. 202231032309 (India)

Advanced Coursework (IIT Bombay) {Till Now}

EE 782: Advanced Topics in Machine Learning [Prof. Amit Sethi]
GNR 650: Advanced Deep Learning for Image Analysis [Prof. Biplab Banerjee]
DH 801: Bio-Statistics In Healthcare [Prof. Saket Choudhary , Prof. Ranjith Padinhateeri]
DH 302: Introduction to Public Health Informatics [Prof. Nirmal Punjabi,Prof. Saket Choudhary]

Education

MS (By Research) in Health AI

IIT Bombay | KCDH | 2025 - Present

B.Tech in Computer Science (AI & ML)

Institute of Engineering and Management [WBUT], Kolkata | CGPA: 9.37