Selected work
Investigated how humans select words during speech by analyzing brain activity using novel machine learning techniques on real neural datasets using Python and MATLAB. Conducted at DiLibERG Lab, the project sits at the intersection of Neuroscience, Computer Science, Cognitive Science, and Psychology, applying computational methods to better understand language and cognition.
Multiplayer browser game built from scratch using Node.js WebSockets. Players join a shared session, submit sentences, and vote on each other's answers across timed rounds. Focused on clean separation between networking and game logic, with a modular architecture that keeps the WebSocket layer decoupled from state management. Handles real-time bidirectional communication using the ws library, with a persistent in-memory game state that survives across rounds without resetting. Implements a full round lifecycle — host assignment, countdown timers, sentence collection, voting phase, and winner calculation — orchestrated with async/await rather than nested callbacks. Player sessions are tracked via a socket map, enabling targeted messaging to individual clients alongside global broadcasts. Built with a clear separation of concerns: server.js owns connections and timing, game.js owns state and rules, and client.js reflect players' actions. Includes a hand-rolled test suite validating core game logic in isolation from the WebSocket layer.
System for retrieving, cleaning, and structuring real quotes from verified sources. Focused on filtering out low-quality or non-notable entries to keep only legitimate, sourced quotes. Processes 500k+ records using Python and Pandas with a structured pipeline for validation, transformation, and deduplication. Uses Wikidata SPARQL API to verify entities and ensure source consistency. Supports querying by author, tag, and category with random sampling from validated datasets. Built with modular components, unit tests, and clear separation of logic, designed to be easily extended into an API with frontend.
Built a Python tool that scrapes and extracts readable text from a given URL, designed for sequential content such as articles or book chapters. The system can follow multiple linked pages based on user input, aggregating and cleaning content by removing unwanted HTML elements and special characters. The extracted text is then formatted into a styled PDF with customizable output options. Implemented request-handling techniques to ensure stable and reliable data extraction while reducing the risk of bot detection.
Developed an interactive Python data visualization project analyzing and comparing the careers of the “Big Three” — Roger Federer, Rafael Nadal, and Novak Djokovic — with emerging players Jannik Sinner and Carlos Alcaraz. Used multiple visualization techniques, including timeline comparisons and statistical graphs, to identify career trends, compare early-career progression, and analyze player strengths and performance patterns.
Built a 48-hour hackathon project at HackTrinity in a team of four: a court hearing summarizer with outcome prediction using NLP and classification, awarded MVP bonus prize. Developed using React and Python with integration of the Perplexity API for intelligent data processing and summarization.
Built a system in C++ that analyzes images to detect pedestrian crossings and visually identify their locations. The model was evaluated against ground truth datasets using precision and recall metrics to measure detection accuracy and assess false positives. The system achieved strong performance on standard and distance-perspective crossing scenarios.
Explored ethical considerations of algorithmic fairness in artificial intelligence by investigating different types of bias and their impact on model outcomes. Used a real-world inspired dataset to train a predictive model estimating the likelihood of student dropout based on multiple socio-academic factors. Analyzed how bias in data and model design can influence predictions and lead to inequitable outcomes.
Developed a C++ application structured using the Model-View-Controller (MVC) design pattern. Built collaboratively in a team of three, the project focused on designing algorithms to efficiently retrieve and process data from monitoring systems tracking both private and public air pollution sources. The application was used to analyze environmental data and identify pollution trends over time.
Developed a full-stack Java application using Spring Boot as part of a three-person team. The system supported two user roles: students and teachers. Students could view their grades across subjects and request video calls with teachers based on availability. Teachers had access to a scheduling calendar, upcoming sessions, and performance analytics related to student grades.