Research on applying XAI methods for Deep Learning architectures widely known to be used in detecting credit card transaction fraud, including CNN and LSTM with attention mechanisms, trained on Sparkov's synthetic dataset. The main contribution lies in the integration and comparative analysis of three Explainable AI methods: SHAP, LIME and Anchors. Research further evaluates the effectiveness of each XAI method based on Faithfulness, Monotonicity, and Completeness metrics.

This project demonstrates the deployment and benchmarking of a virtual computer cluster on Microsoft Azure. The system consists of three Ubuntu 22.04 nodes (one headnode and two compute nodes) connected through a shared virtual network (HPC-vnet), using OpenMPI for parallel processing.

A Python tool designed to automatically scrape contents for students from Year 1 to Year 10-11, 11+, and Year 12-13 (A-levels). The scraper extracts quizzes and practice exams from various educational websites using Selenium, Beautiful Soup, LLM models (OpenAI API) and OCR (MathPix API).

Using Prompting techniques to develop a chatbot system for helping customer questions. The chatbot is designed to serve as a customer service assistant for the Learnspot website, a platform that aids students in preparing for tests and connecting with tutors. Using prompting techniques to assist with customer inquiries and prevent hallucinations. The assistant is programmed to respond in a friendly and helpful manner, providing concise answers and asking relevant follow-up.

Our solution for the McGill-FIAM Asset Management Hackathon. The hackathon challenges participants to apply machine learning (ML) techniques and data-driven approaches to design innovative portfolio trading strategies. At LYTA Strategy Analytics, we developed a mixed long-short investment strategy using advanced ML techniques, achieving significant performance gains over traditional market benchmarks.

Collaborating with a PhD candidate on a significant project involving the analysis of an eye-tracking dataset. My main role was to assist in deriving valuable insights from the dataset by categorizing fixations into "First Pass" (initial read) or "Second Pass" (subsequent read) sequences. Realizing unusual points while analyzing the data, I developed an Anomaly Detection mechanism for the dataset where using the Local Outlier Factor (LOF) to identify and remove outliers, establish criterias for accurately determining when a reader transitioned to the next line of text. These contributions notably improved the robustness and accuracy of the data analysis process.

To help rescue crews and retrieve the lost passengers in 2912. I am challenged to predict which passengers were transported by the anomaly using records recovered from the spaceship's damaged computer system. The dataset contains 12 features, including PassengerId, HomePlanet, CryoSleep, Cabin, Destination, Age, and more. I used XGBoost and ANN to predict the target variable (Transported) with an accuracy of 0.8.

Develop a collaborative-filtering recommender system that serves 10 personalized exercise suggestions, retraining in mini-batches from live user ratings. Deployed on AWS (EC2 for model serving, S3 for data storage, RDS for training data), the solution was built with strong PLESI focus—privacy, legal/ethical, security, and gender-inclusive design.