A full-stack pipeline to predict U.S. traffic accident severity

• •Built a full-stack pipeline to predict U.S. traffic accident severity using BigQuery, PyTorch neural networks, and Plotly Dash on Google App Engine.
• •Engineered over 20 weather, road, and time-based features and trained classification models to distinguish severity levels with high accuracy.
• •Used NLP to classify potential causes from accident descriptions and visualized findings through interactive dashboards.

Human activity recognition using time and frequency domain features from motion sensors

• •Segmented motion sensor data using a sliding window approach and extracted time and frequency domain features including mean, std, skewness, and FFT components.
• •Trained and evaluated multiple models including KNN, SVM, and Random Forest, achieving over 82% accuracy in LOSO (Leave-One-Subject-Out) cross-validation.
• •Demonstrated the effectiveness of classical ML over deep learning for lightweight and interpretable human activity classification.

A comprehensive platform for university club management

• •Developed and deployed a club management platform using React for the front-end and Node.js for the backend, implementing dashboard navigation, dark mode, and club creation features to enhance user experience.
• •Collaborated with a cross-functional team to build a real-time Chat Room with @mention functionality, integrating Node.js backend queries for seamless communication.
• •Engineered secure authentication by designing React components and Node.js middleware for sign-up functionality, private routes, and user restrictions.

CNN-based food image classification using ResNet50 and Food11 Dataset

• •Built a convolutional neural network to classify food images across multiple categories using transfer learning with pre-trained ResNet architecture.
• •Implemented data augmentation techniques and fine-tuning strategies to improve model accuracy.
• •Created visualizations to analyze model performance and feature importance.

Python-based productivity tool for tracking and managing application usage time

• •Designed a smart time-management application that monitors and tracks time spent on various applications to improve productivity.
• •Implemented features including a focus mode, calendar view for usage history, and detailed statistics visualizations.
• •Created a user-friendly interface with customizable settings for application names, autosave functionality, and Pomodoro timer configuration.

Machine learning model for predicting potential BRAF inhibitors in cancer drug discovery

• •Developed a machine learning pipeline to identify potential small molecule inhibitors targeting BRAF protein mutations for cancer treatment.
• •Implemented data preprocessing, model training, and evaluation workflows to ensure accurate predictions for early-stage drug discovery.
• •Created a Flask-based web application interface for researchers to easily interact with the predictor and analyze results.