I am a MSci Computer Science student studying at King's College London specialising in AI and Machine Learning!
Contact me via email at matteoafortier@gmail.com or connect with me on LinkedIn! 👥💬
Check out my favourite personal projects!
Steam Games' Genre Multi-Label Classifier (Deep Learning)
- Ingested 40,000 games to produce a dataset of 10,000 video game screenshots and 13 labels.
- Developed custom tensor generators to support multi-label data and optimisations such as dataset cache-ing.
- Implemented image augmentation techniques to reduce overfitting, improving model macro-f1 score by 0.3.
- Developed a tailored macro-f1 metric and soft-macro-f1 loss function to score and train multi-label classification.
- Modelled a CNN video-game genre labeller using TensorFlow and VGG-16 achieving a macro-f1 score of 0.54.
Recipe Recommender (NLP, Unsupervised Learning)
- Ingested and pre-processed 1 million recipes to extract ingredients using NLTK, SpaCy, and Google Compute Platform.
- Designed a topic modelling pipeline using TFIDF vectorizer and NMF to produce 20-100 intuitive topics.
- Deployed a web-app using Streamlit that recommends the top 15 most similar recipes given an inputted/selected recipe.
Steam Game Success Classifier (Classification)
- Produced a dataset of 40,000+ rows with 29 features from disparate sources to use for classification.
- Automated tuning of 200 combinations of hyper-parameters using GridSearchCV.
- Evaluated the performance of classification models such as XGBoost to achieve AUC of 0.922.
Twitch Stats Dashboard (Data Engineering)
- Automated data collection from Twitch API to a MongoDB Atlas instance every 30 minutes.
- Designed 8 interactive (twitch) data visualisations using Plotly and Dash.
- Deployed a web-app on Heroku to allow for quick and easy access to live and historical data about Twitch and Twitch Partners.
University Course Satisfaction Predictor (Regression)
Exploring the Relation Between Yankees Home Games and MTA Subway Stations (EDA)