Projects developed during my studies at UAB.
Deep learning project for automatic image caption generation using CNN–LSTM architectures and Bahdanau Attention. Trained and evaluated on Flickr8k and Flickr30k using ResNet-152 as encoder, LSTM decoders, GloVe embeddings, and beam search.
End-to-end NLP project for sentiment classification of tweets. Features a preprocessing pipeline for text cleaning, lemmatisation, and noise removal. Benchmarks multiple ML approaches including SVM, LightGBM, and voting ensembles, with a deployed FastAPI web application.
Machine learning project developed for the FME Datathon 2025, focused on predicting product-level demand for Mango. We worked with a dataset of over 90,000 training samples and product attributes such as family, silhouette, season, fabric, colour and image embeddings. The solution combined LightGBM, CatBoost, Ridge Regression and ensemble techniques to build a more accurate, stable and interpretable forecasting system.