A collection of my personal projects
Enhancing Credit Card Fraud Detection through Machine Learning and Deep Learning
Developed an advanced credit card fraud detection system using machine learning and deep learning to analyze real-life transaction data and identify fraudulent activities. Technologies used include Python, Scikit-learn, TensorFlow, XGBoost, Pandas, and NumPy.
Implemented and compared models such as CNN, Autoencoder, KNN, Random Forest, and XGBoost, achieving 99.97% accuracy and a Matthews Correlation Coefficient of 0.91007 with XGBoost. Addressed challenges like computational limits with GPU-accelerated algorithms and overfitting through regularization techniques.
Raft Consensus Algorithm Implementation
Implemented the Raft consensus algorithm in Python, applying distributed systems concepts and protocols. Technologies used include Python, Threading, RPC, and Distributed Systems.
Developed core Raft components like state management, log entry structures, and role-based behavior. Implemented a robust election process with vote requests and append entries, using threading for concurrent operations and managing election timers. Designed serialization methods for inter-node communication.
Created a `Node` class to manage states and role transitions, with synchronization mechanisms for thread-safe operations. Developed a role-based system (Follower, Leader, Candidate) for node behavior.
Gained insights into distributed consensus complexities and identified improvements in error handling and network partition management. Optimized timing mechanisms for varied operational environments.
PawClock
PawClock is an iOS app designed to help pet owners manage daily activities and meals for their pets, with an interactive game to engage both pets and owners.
Features: Schedule and track pet meals and activities, receive notifications for upcoming events, and play an interactive game.
Technologies: SwiftUI for UI building, SpriteKit for the game component, and Local Notifications for reminders.
Buy Earth a Coffee Application
Developed a full-stack app using Next.js for user profile management and cryptocurrency donations. Built the backend with Node.js and MongoDB, utilizing Mongoose for schema creation. Integrated AWS S3 for secure file storage, and implemented user authentication with Next-Auth. Employed Axios for transactions with the Cryptomus payment gateway and used React to enhance the UI with custom components and state management.
NuGraph: a Graph Neural Network (GNN) for neutrino physics event reconstruction (Partnering with Fermi Lab)
Advanced NuGraph3 GNN architecture, optimizing data aggregation and message-passing for enhanced event-level predictions. Implemented a sawtooth mechanism for sequential node embedding updates, refining model accuracy and performance. Applied residual connections in NuGraph3, boosting robustness and feature refinement across iterative message-passing. Streamlined data pipelines with Python, enabling efficient data handling and supporting advanced analytical capabilities.
Genomic Annotations Service
Led the design and development of a Genomic Annotations Service, a web service for gene data analysis utilizing Flask, Globus, and AWS cloud technologies including S3, EC2, SQS/SNS, DynamoDB, Lambda, and Step Machines. Built RESTful APIs with JavaScript and Python, managing data workflows and service integration to provide a robust user experience.