I'm Devraj Raghuvanshi.
A Data Scientist & Machine Learning Engineer

This was a group project completed as the final project for Harvard's AC215 (AIOps) course. We built Spatially, an AI-powered tool that helps make sense of complex zoning regulations and urban development data by letting users ask questions directly and see answers grounded in official documents like zoning ordinances, maps, census data, and real development plans. The system pulls together messy, large-scale public data and uses a RAG-based setup to surface relevant information without hallucinating.

This project explored human activity recognition from wearable accelerometer data, with an emphasis on building models that generalize across individuals without requiring user-specific retraining. A structured, end-to-end machine learning pipeline was created, covering data preprocessing, windowing, feature engineering, model training & evaluation. Several machine learning models were implemented and systematically compared using evaluation strategies designed to handle user variability and class imbalance.