Subhrajyoti Roy
Data Scientist
Pune,Maharashtra
Summary
Implemented fraud detection strategies and risk profiling methodologies to enhance underwriting accuracy and mitigate financial losses. Worked on end-to-end development of models from data analysis and feature engineering to deployment in production while working with highly skewed data
Managed comprehensive project delivery for two initiatives, focusing on stakeholder satisfaction and operational efficiency.
Overview
2
2
years of professional experience
1
1
Certification
Work History
Junior Data Scientist
Star Health and Allied Insurance
07.2024 - Current
- Developed and optimized additional rule sets within an existing rule-based fraud detection framework by fine-tuning thresholds through cross-functional collaboration, improving overall precision by 30% and delivering $3M in quarterly cost savings.
- Built a machine learning–based customer risk profiling model using historical customer data, doubling precision and improving recall by 1.5× over existing benchmarks, resulting in $2M in quarterly savings.
- Designed and deployed a fraud propensity prediction model leveraging historical claims data and third-party industry datasets (CIBIL, CRIF, PayU, GeoIQ), achieving 82% accuracy and identifying 74% of fraud cases within the top 4 risk deciles.
- Developed a GenA-driven automated document classification pipeline using Azure OpenAI to categorize incoming proposal documents into medical and non-medical cases, improving overall turnaround time (TAT) by 20%.
- Engineered a scalable NLP-based de-duplication system to detect fraudulent or duplicate customer records across a portfolio of 9M+ customers, achieving 98% clustering accuracy.
- Implemented a deep learning–based document forgery detection model using PyTorch, incorporating image-level anomaly detection and orientation correction techniques, achieving 70% precision in identifying digitally altered claim documents.
Education
PGD - Statistical Methods & Analytics
Indian Statistical Institute
Kolkata01.2024
MSc. - Physics
Presidency University
Kolkata01.2023
BSc. - Physics
Presidency University
Kolkata01.2021
Skills
Python/R
Certification
- SQL - MySQL for Data Analytics and Business Intelligence, 07/01/24 - 09/30/24
- Hackerrank Python (Basic), 10/01/24
- Hackerrank SQL (Basic), 02/01/25
Languages
Python
Java
SQL
R
Fortran
PGQL
Project
Deepfake Detection with ViT-CNN, Trained a CNN on 250 GB of video data to detect deepfakes with 95% accuracy. Integrated Vision Transformers to mitigate catastrophic forgetting, with applications in cyber forensics and content moderation. Research Publications, One-dimensional quench dynamics in an optical lattice: Sine-Gordon and Bose-Hubbard descriptions. Phases and coherence of strongly interacting finite bosonic systems in shallow optical lattice.
Personal Information
Active Learner with research skills and analytical mindset working in a fast paced and evolving field of data science. Beyond office hours you can find me playing a ukulele, sketching or travelling.
Timeline
Junior Data Scientist
Star Health and Allied Insurance
07.2024 - Current
PGD - Statistical Methods & Analytics
Indian Statistical Institute
MSc. - Physics
Presidency University
BSc. - Physics
Presidency University