Data Science
Data science projects, ML experiments, and Python programs
Some of these projects — the RAG Educational Assistant and the Datathon — were developed as part of my Intensive Program in Data Science and Master's in Deep Learning & Generative AI at Datamecum. The rest are personal projects.
Manifold HyperConnections for Computer Vision (in progress)
Computer Vision · Research · PyTorch
Replicating and extending Manifold HyperConnections (arXiv:2512.24880) for computer vision — original work exploring how mHC applies beyond the settings covered in the paper. Currently benchmarked on ImageNet and CIFAR-100, with a fully configurable CLI built on argparse to run and reproduce experiments.
CON(e)VOLUTION — From LeNet to Vision Transformers
Computer Vision · PyTorch
10 CNN architectures implemented from scratch in PyTorch — BaselineCNN, LeNet, LeNetMod, LeNetModNorm, LeNetModNorm2, AlexNet, GoogLeNet, ResNet, DenseNet, and ViT — all benchmarked on the NEU Surface Defects dataset. Each model shares a modular trainer and CLI interface, making it straightforward to run and compare experiments. Paired with a deep-dive Medium article walking through the full evolution from early CNNs to Vision Transformers.
Datamecum Datathon — 1st Place
Machine Learning · Competition
1st place in a Kaggle-style binary classification datathon. Built an ensemble of Random Forest and XGBoost (AUC 0.9851) — work included EDA, self-organising maps, correlation analysis, and iterative ensemble tuning.
RAG-Driven Educational Assistant
Master's Dissertation — Deep Learning & Generative AI
Turned 364h of video content (91 lessons) into an instantly queryable knowledge base — Whisper for transcription, LangChain + ChromaDB for retrieval, GPT-4o-mini and Claude-3-Haiku for responses. Deployed as a Flask web app with a CLI, used by real Datamecum students and benchmarked against them on a gamified leaderboard.
Ryanair Timecapsule
Data Collection · Open Source
Built a daily flight price tracker by reverse-engineering Ryanair's Fare-Finder and Booking APIs — designed to feed ML models for flight price forecasting. Supports flexible querying across markets, date ranges, and passenger configurations. Released as an open-source Python package with tests, CLI, and a proper src layout.
Python Exercises
Throughout my journey of learning Python, I have created several interesting projects. My learning started with the Udemy course #100DaysOfCode and for the last 20 projects I worked independently. You can access all of them on my GitHub.
Morse Converter
Command-line · Python
A command-line program for encoding and decoding text to and from Morse code. Includes a test suite — the first time I wrote tests to verify the correctness of my own code.
Tic Tac Toe
Command-line · Python · OOP
Classic two-player game implemented as a command-line app. Built around a single class containing all game logic — a great exercise in OOP and working with 2D arrays.
Typing Thunder
GUI App · TKinter · Python
A speed-typing app where the user types for one minute and gets their character count at the end. The trickiest part was implementing the countdown timer correctly within TKinter's event loop.
Weever Watermark
GUI App · TKinter · PIL · Python
A GUI tool that automatically places a watermark at a consistent position and scale on any image — something I used to do manually in Photoshop as a photographer. Built with TKinter and PIL.
MochaMaps
Web App · Flask · SQLAlchemy · SQLite · Jinja2
A web app for browsing, adding, and deleting coffee shop listings backed by a SQLite database and REST API. My first project using Flask, SQLAlchemy, and Bootstrap template inheritance together.