I recently completed an MSc in Data Science & Analytics. I build Python applications and work on NLP experiments, especially with multilingual, noisy, or limited data. I enjoy both sides of the process: understanding how a model behaves and building the software that makes it useful.
My research interests include Moroccan Darija, code-switched text, and model evaluation.
Currently seeking a full-time junior AI, Python, or software engineering role. I'm also interested in funded NLP/ML research positions and PhD opportunities.
I'm currently building ApplyLens AI, a workspace for reviewing academic opportunities against candidate evidence and organizing applications.
|
Low-resource NLP |
Transformers |
Python & FastAPI |
- Master's research: Character-aware XLM-R for depression-severity text classification in Moroccan Darija, comparing character embedding injection and gated adapters for Arabizi and code-switched text.
- Language and speech projects: emotion-aware summarization, speaker verification, and adversarial evaluation.
- Other academic work: ontology-based reasoning and interpretable decision systems.
I'm working on making my experiments easier to reproduce and documenting their methods, results, and limitations clearly.
|
A workspace that helps Master's and PhD candidates review eligibility evidence, identify missing information, compare opportunities, and track application tasks. Includes automated tests, CI configuration, and a synthetic evaluation. Default analysis uses deterministic rules and lexical retrieval; external embeddings and pgvector retrieval are optional.
|
Depression-severity text classification in Moroccan Darija and Arabizi, comparing XLM-R with character-aware embeddings and gated adapters. Includes model architectures, documented test results, and a reproducibility audit.
|
|
A comparative study of ECAPA-TDNN and WavLM speaker embeddings under gradient-based perturbations, evaluated through cosine similarity and equal error rate.
|
A collaborative academic project combining RDF/OWL ontologies, Python rules, and SPARQL in a Streamlit interface for simulated banking transactions.
|



