ayoub chabi

I’m an AI/ML Engineer specializing in Arabic NLP, LLM-based systems, and Explainable AI. In my Master’s thesis, I built an AI-generated text detection model across temperature settings using AraBERT and CamelBERT, reaching an F1-score of 0.91 on 8,019 samples with Qwen 2.5 7B data.

I focus on reliable behavior and interpretable decisions. I designed cross-temperature evaluation to study asymmetric generalization failure and applied explainability techniques such as SHAP, LIME, Attention, and Integrated Gradients across multiple distortion types.

I also build practical LLM systems with RAG and orchestration. My NLP chatbot project implements a PDF ingestion pipeline with text chunking and ChromaDB storage, then deploys Google Gemma 2B via Ollama for local CPU inference, integrated with LangChain and a Streamlit web UI.

Beyond that, I develop NLP solutions for real-world dialect and domain needs, like a search engine for Algerian Darja with Arabizi–Arabic conversion, sentiment lexicon polarity inversion, and a positional inverted index using TF-IDF cosine similarity. I enjoy end-to-end ML systems—clean data, robust modeling, and usable interfaces—using tools like Python, PyTorch, HuggingFace Transformers, and Git.