My Projects
Work organized by university semester, plus personal projects built along the way.
Blackjack AI
A Python OOP implementation of Blackjack featuring an AI player that uses probability calculations to decide whether to hit or stand. Includes matplotlib visualizations to analyse win rates and strategy performance across thousands of simulated games.
HARDS
Autonomous Robot Delivery Service — a simulation of drones and legged robots navigating a city grid using waypoints. Built with trigonometry and numpy for pathfinding, with full OOP class hierarchies for different robot types and delivery scenarios.
Submarine Sonar Simulation
A probability simulation of a submarine detection game using sonar searches and dice roll mechanics to explore theoretical vs experimental probability. Two variants implemented: single-shot and double-shot per round.
KIUA1011: Intro to AI — Final Project
Group project for the Intro to AI course at INN (KIUA1011). Four exercises covering: conceptualising AI tools for student orientation at HINN, training an image classifier on 210 manually curated and licensed images (Cars, Cats, Dogs) using Google Teachable Machine, and written reflections on AI ethics and responsible use in education.
This Portfolio Website
A full-stack portfolio built with Next.js and deployed on Vercel. Features guest/admin password protection, a dynamic admin panel for managing projects, dark mode, and Telegram notifications on contact form submissions.
Crypto Bros — Bitcoin Trading Simulator
A menu-driven Bitcoin trading simulator built in Python. Loads real historical BTC/USD price data from Kaggle, calculates SMA 20 and SMA 50 indicators, generates buy/sell crossover signals, and compares three trading strategies: SMA Crossover, Price vs SMA 50, and Buy & Hold. Features an animated price chart and bullish/bearish sentiment analysis.
Solo Leveling GPT-2 Chatbot
Fine-tuned a pretrained GPT-2 model on a custom 44-pair Q&A dataset about the Solo Leveling universe. Modified the base project to load training data from CSV instead of hardcoded pairs, then trained the model to answer lore questions in an interactive terminal chatbot.
Evolutionary Jumper Simulation
A physics-based simulation where virtual agents evolve over 80 generations to improve their jumping ability. Each individual has a genome controlling jump angle, force, and body dimensions. The best performers are selected each generation, cloned and mutated — mimicking natural selection.
Tabula Rasa — Sisyphus Reinforcement Learning
A reinforcement learning agent that learns to walk, push a boulder, and climb a slope — from scratch. Inspired by Camus and the Sisyphus myth. Trained with PPO over 60 million steps total across three transfer learning phases using a custom Gymnasium environment.
In progress
In progress
In progress
In progress
In progress
In progress
In progress
In progress