M.Tech in Data Science from ISI Kolkata. Forecasting models in production, Graph Transformer research, and large-scale data on GCP behind Walmart's planning and forecasting.
M.Tech in Data Science from ISI Kolkata (AIR 11), with a dissertation on reducing attention complexity in Graph Transformers. Built and deployed an electricity-trading forecast model at Tata Power and improved a payment-default model at Allstate.
At Deloitte, on a Walmart engagement, I build the Spark-Scala data products on GCP that feed Walmart's enterprise planning and forecasting. I also teach Probability, Statistics and Machine Learning for GATE DA, and build full-stack apps — for example an AI-powered notebook.
My journey across machine learning, production systems and teaching.
Strong foundation in computer science and data science.
What I use at work, and what I'm building with now.
ML models, research, and the data systems behind them.
ARIMA and XGBoost forecasts for the Haldia power plant: generation from PI System sensor and Tata Steel data, sales from IEX market data, for day-ahead trading. Built and deployed.
Reduced Graph Transformer attention from O(N²) to O(K²) by partitioning the graph into K subgraphs (METIS) and attending over a supergraph of GCN + Laplacian-embedding subgraph tokens.
Improved an XGBoost payment-default prediction model using geographic feature engineering and hyperparameter tuning.
Spark-Scala data products on GCP — sales and margin over rolling windows, order-substitution analytics — that feed Walmart's daily enterprise planning and forecasting.
AI-powered note-taking app with rich editor, code blocks, math rendering, drawing canvas, PDF export. Integrates Gemini and Groq for AI writing. Deployed as PWA.
End-to-end automated Power BI dashboards for HR, customer operations, and engineering teams. REST APIs with FastAPI and Node.js serving 10+ dashboards.