NLP from first principles ↗
Persian tokenizers, N-gram language models, text classifiers, and word embeddings implemented from scratch; a FastText–MLP news classifier.
Video category classification ↗
An SQL-to-model pipeline combining sentence embeddings and engagement features in a multi-branch MLP.
3D sound source localization ↗
Learning-based localization using audio timing and spectral features. Published at ICRoM 2024. See the publication for results and citation.