Build batch processing pipelines to do behavior analytics.

Unified ML pipeline for user behavior prediction and personalized nudge generation using 5 integrated modules

Full-stack e-learning platform featuring a FastAPI backend and a React/Tailwind (Vite) frontend. It tracks course and resource progress in real time, auto-generates implicit ratings from viewing behavior, and delivers personalized recommendations via an SVD-based machine learning pipeline, with explicit ratings stored and displayed for users.

Real-time social engagement pipeline that ingests user behavior streams, infers psychological state (bored, calm, stressed), and updates live user profiles for feed personalization.

A data analysis pipeline that establishes personalized behavioral baselines from daily metrics and identifies meaningful deviations while filtering out noise.

Built an end-to-end recommen dation pipeline modeling user behavior sequences and serving personalized item predictions with a simple front-end.

In the world of personalization, static demographic segmentation no longer cuts it. So I built an end-to-end unsupervised pipeline that uses autoencoders + clustering algorithms (KMeans/DBSCAN) to discover hidden behavioral patterns in real-world user activity data.

Developed a Python-based analytics pipeline using pandas, seaborn, scikit-learn, and XGBoost to analyze 39K+ email sends and user engagement behaviors. Unified disparate logs, created a hierarchical engagement metric, performed cohort/segment analysis, and quantified the impact of variables like subject format, personalization, and timing.

This project develops a machine learning pipeline to classify user behavior as "active" or "inactive" based on their in-app interactions. By analyzing patterns in session duration, clicks, and device information, we can predict user engagement, which is crucial for personalization, user retention strategies, and business decision-making.

AI-powered assistant for personal finance and debt management. Uses LLMs and RAG pipeline to deliver personalized advice based on real guides and behavioral principles. Built with modern NLP tools and a user-friendly web interface.

A Virtual Lab platform for EdTech that simulates STEM experiments while capturing student interaction data. The project designs a data architecture and ETL pipeline to transform learning behavior into actionable insights, supporting personalized learning and data-driven educational decisions.

A personalized hybrid recommendation pipeline that combines collaborative filtering (ALS), co-occurrence analysis, LightGBM ranking, and behavioral signals to deliver user-aware product recommendations. It incorporates user profiles, transaction history, recency weighting, and clustering for high-quality, diverse, and interpretable recommendations.

A sophisticated end-to-end Machine Learning pipeline designed to analyze traveler behavior and predict high-satisfaction destinations. This project leverages an XGBoost classification model to rank attractions based on a hybrid of historical popularity and personalized user preference ratios.

This repository contains an end-to-end Content Recommendation System built using Python and SQL-based data pipelines. The project focuses on designing a scalable and production-ready recommendation engine that personalizes content based on user interaction and consumption behavior.

The Recommendation System implements an end-to-end customer analytics pipeline for an e-commerce platform. It combines RFM (Recency, Frequency, Monetary) analysis to segment customers based on their purchasing behavior with a Content-Based Collaborative Filtering approach to generate personalized product recommendations.

Engineered a scalable, real-time data streaming and analytics pipeline for customer clickstream data, leveraging Apache Kafka, Spark Streaming, AWS Kinesis, S3, and Redshift. This system processes high-volume events to deliver immediate insights into user behavior for dynamic dashboards and personalized experiences.

E-commerce companies often struggle with understanding customer behavior across multiple platforms (app, web, call center). This project builds a scalable data pipeline that unifies, cleans, and analyzes raw event data from multiple sources — enabling personalized product recommendations, churn detection, and regional trend analysis.

In today’s fast-paced e-commerce market, businesses must deeply understand customer behavior to deliver personalized experiences. The Shopper Spectrum project implements a comprehensive machine learning pipeline that unifies customer segmentation and product recommendations to support smarter decision-making and improved customer engagement.

[.

Home surveillance systems available today notify homeowners of any motion detected by the cameras. This behavior leads to mostly false positives and annoying notifications. Worse, these systems suffer from the “one size fits all” approach and do not offer any ability to provide personalized feedback and make the notifications only appear for unknown persons and objects. The personalization can be based on the unique features of a home’s landscape, location and known persons and pets. Our team intends to address this gap. We are proposing to develop the building blocks for a smart home surveillance system. The system will use video streams from off-the-shelf outdoor IP cameras. It will analyze and interpret the content of the streams and employ feedback mechanisms to generate customizable notifications for the homeowner. We plan to implement an end to end pipeline that can detect objects and faces in the data streams in order to recognize unfamiliar faces, identify anomalous situations and notify the homeowner about the anomaly. The feedback loop will enable the system to improve its capabilities over time and generate notifications that are personalized for each homeowner.