An eight-variate Hawkes process calibrated by maximum likelihood to a full day of Apple (AAPL) TAQ top-of-book data. The calibration reveals a strikingly stable near-critical spectral radius of 0.95 across the trading day, and a multitype Galton-Watson cluster decomposition showing that around 95% of top-of-book events are endogenously triggered rather than driven by outside information.
A Trustworthy Machine Learning approach built using Bayesian Deep Learning and Vision Transformers (ViTs) to process medical OCT scans. This system explicitly decouples aleatoric and epistemic uncertainty to prevent overconfident automated diagnoses and reliably route ambiguous cases in clinical triage queues.
A surrogate model built using Physics-Informed Neural Networks (PINNs) regularised by the Navier-Stokes equations to optimise turbine yaw angles. This digital twin achieves a power gain of up to 4.65% under realistic wind conditions for a Dogger Bank-inspired offshore wind farm structure.
A multi-layer perceptron (MLP) model for digit classification using the MNIST dataset. The model is trained using the backpropagation algorithm and achieves high accuracy in classifying handwritten digits. The project demonstrates the implementation of a neural network for image recognition tasks using pytorch.
A vision transformer model for image classification using the CIFAR-10 dataset. The model is based on the transformer architecture (From the 'Attention is all you need' paper) and achieves high accuracy in classifying images into different classes. The project is developed using pytorch.
This project deploys and trains a continuos pipeline using ZenML(an mlops framework) to predict the housing prices. The pipeline includes data ingestion, preprocessing, feature engineering, model training, and evaluation. The project demonstrates the implementation of multiple regression models for predicting housing prices.
A conversational chat interface for exploring global internet trends and outages using Cloudflare's Radar API and AI. Users can ask questions about internet outages, traffic trends, and network performance, receiving AI-powered responses based on real-time data.
A robust backend for an e-commerce website using Node.js, Express, and MongoDB. It manages user authentication, product inventory, order processing, and etc. The backend ensures efficient data handling, supporting seamless e-commerce operations. The code is divided into 4 main modules: Models, Routes, Controllers, and Middleware.
A Java web application for managing customer information using MySQL. The application allows users to perform CRUD operations on customer data, including adding, updating, and deleting customer records. The project demonstrates the development of a web application using Java technologies. The project is developed using Java, JSP, Servlets, and MySQL. The project was built by following the entire software development life cycle.
A Tetris game developed using JavaFX. The game features a graphical user interface with different levels of difficulty and scoring. The project demonstrates the implementation of a classic arcade game using Java technologies. The project is developed using Java, JavaFX, and CSS. It is built using maven.
A web portfolio developed using Astro, TailwindCSS, and React. The portfolio showcases projects, blog posts, and contact information. The project demonstrates the implementation of a responsive web portfolio using modern web technologies. There are animations and transitions to make the website more interactive.