About

Hello and welcome! My name is Bibhabasu Mohapatra, and I am a passionate Deep Learning enthusiast dedicated to advancing the field of artificial intelligence. Currently working as a Data Scientist at Microsoft, previously working as an Engineer(Computer Vision) at Aira Matrix. I'm deeply committed to pushing the boundaries of AI research.

I'm particularly interested in topics related to self-supervised learning, multi-modal learning, and Representations. These areas fascinate me because of their potential to revolutionize AI and address complex challenges in various domains. Open for Research Opportunities in Academia.

Education

International Institute of Information Technology, Bhubaneswar, India
Bachelor of Technology (Computer Science & Engineering)
August 2019 - July 2023
Cummilative GPA: 8.05

Experience

Microsoft, Bangalore, India
Data and Applied Scientist (June 2024 – Present)
• Responsible for demand planning of compute resources required across Azure data centers globally, encompassing both short-term (4-5 months) and long-term (2-3 years) forecasts
• Developed a Hierarchical Reconciliation Module to enhance demand forecasting for all resource types across various geographic levels

Aira Matrix (Sun Pharma Company), Mumbai, India
Engineer(Computer Vision) (June 2023 - June 2024)
• Implemented a Hierarchical Image Vision Transformer(HIPT), using DINO, and trained weakly supervised classifiers achieving 90% AUC and 88% accuracy
• Developed patient-level risk stratification solution using multi-parametric MRI, achieving 84% test AUC
• Generated 3D reconstruction of segmented tumor and prostate from 2D WSIs and MR Images

Aira Matrix (Sun Pharma Company), Mumbai, India
Intern(Computer Vision) (Feb 2023 – June 2023)
• Developed Tumor Lesion Segmentation models for Prostate MRI data
• Acquired expertise in Prostate Histopathology and Gleason Grading

Polynomial.ai, Bangalore, India
Machine Learning Intern (May 2022 – July 2022)
• Contributed to migration of Product Interaction Studio from Dialogflow to Rasa
• Achieved 3-4% optimization in performance for entity and regex extraction

Projects

HuBMAP + HPA - Hacking the Human Body
• Won Bronze Medal in competition to identify functional tissue units (FTUs)
• Employed Segformer for Semi-Supervised Training on large images

U.S. Patent Phrase-to-Phrase Matching
• Ranked in top 20% (366/1889) using DeBERTaV3 for semantic similarity

Transformer Playlist: A from scratch Implementation
• Created educational content on Vision Transformer implementation
• Demonstrated advanced PyTorch techniques and attention visualization

Publications

Fused Deep Learning Paradigm for Predicting o6-Methylguanine-DNA Methyltransferase Genotype
• Conducted radiomics feature extraction on multiple MRI types
• Achieved AUC ROC score of 0.61 using Nvidia MONAI and ResNet 3D

Achievements

• Bronze medalist (80/1245) in HuBMAP + HPA Competition
• Top 1% on Kaggle Notebook Expert Leaderboard (1266/222,350)
• Top 20% in US Patent Phrase Matching Competition
• Top contributor in Hacktoberfest for PyTorch Lightning
• Certified in NPTEL Computer Vision and Image Processing

Skills

Software: Python, PyTorch (extensive), OpenSlide(extensive), blender, SimpleITK(extensive), Scipy, QuPath, WandB
Specialization: Computer Vision(extensive), NLP, Self-Supervised training, Representation Learning