My Portfolio

RECOMMENDATION ENGINE

This project implements a two-phase recommendation system for a E-commerce website. The first phase, candidate generation, efficiently narrows down a vast pool of potential recommendations. The second phase, ranking, scores and orders the candidates based on user preferences and item characteristics, delivering the most relevant recommendations.

AI Chatbot Agent

The AI Chatbot Agent is an advanced tool designed to improve user interactions with e-commerce platforms. With its powerful capabilities, it allows users to:

  • Access detailed product information.
  • View images of products.
  • Place orders directly on the website.

Using Retrieval-Augmented Generation (RAG), the chatbot combines intelligent information retrieval with seamless e-commerce functionalities, offering a smooth and user-friendly shopping experience.

FITNESS EXCERCISE TRACKER

The Fitness Exercise Tracker is an innovative tool designed to monitor and analyze physical activities in real time. Using data from accelerometers and gyroscopes, this tracker accurately identifies the type of exercise being performed. Whether it's bench press, deadlift, row, squat, overhead press,and also detect rest periods when the user is not performing any exercise.

LOW COST SENSOR CALIBRATION

This project focuses on improving the accuracy of low-cost air quality sensors by calibrating them against high-precision FEM (Federal Equivalent Method) sensors using machine learning. By leveraging supervised learning techniques, the model learns to correct the biases and inaccuracies of low-cost sensors, ensuring more reliable PM2.5 measurements. The solution is deployed on AWS, enabling scalable and cost-effective real-time calibration. This approach significantly reduces the reliance on expensive sensors while maintaining measurement accuracy, making air quality monitoring more accessible and affordable.

Customer Lifetime Value (CLV) & Next Purchase Prediction

This project predicts Customer Lifetime Value (CLV) and estimates when a customer will make their next purchase using machine learning. By analyzing historical transaction data, the model identifies spending patterns, purchase frequency, and customer engagement to forecast long-term value. Time-to-next-purchase prediction helps businesses optimize marketing strategies, improve retention, and personalize offers.

The Blog

Balancing Accuracy and Importance: Why Low Average Test Error Is not Good Enough

Building solutions in machine learning is hardly a one-size-fits-all approach. The field has traditionally focused on building machine learning models with unmatched accuracy, so celebrating the achievement of a low test error is very common. In many real-world applications, low-average test errors fall short of addressing the solution’s significance. Conducting thorough research is crucial to avoid jumping to conclusions based solely on the model's accuracy.

MACHINE LEARNING MODEL DEPLOYMENT PATTERNS

Imagine building machine learning models that can accurately predict customer behaviours, automate decision-making, or enhance the user experience. While the model development process is fascinating, the real magic happens when these models seamlessly and effectively transition from juypter notebooks to practical, real-world applications.