I build end-to-end AI products: from document ingestion and retrieval to model serving, evaluation, and production deployment. My work focuses on agentic RAG systems, Arabic document intelligence, and practical MLOps.
Currently, I am building agentic workflows for Arabic legal-document preparation and research. Previously, I worked on production OCR/VLM serving and NLP systems across cloud and Kubernetes environments.
| Project | What it demonstrates | Stack |
|---|---|---|
| resume-tailor | An AI resume tailor that rewrites resumes per job description without inventing experience, backed by an LLM pipeline and PDF rendering. | FastAPI, React, LangGraph, OpenAI SDK, LaTeX |
| online_khatt | Arabic online-handwriting recognition research and model fine-tuning workflows. | Python, deep learning, OCR |
| AtaAI | An Arabic AI learning companion delivered as a usable application. | Python, Streamlit, Docker |
| Mizan | A bilingual personal-finance product with a documented full-stack deployment. | React, FastAPI, PostgreSQL, Docker |
I am preparing a public, sanitized reference implementation for agentic RAG and document-processing workflows. It will demonstrate the production patterns behind my recent GenAI work without exposing client or employer code.
- GenAI & ML: LangChain, LangGraph, RAG, PyTorch, Hugging Face, VLMs, OCR
- Production ML: Ray Serve, Kubernetes, Docker, model serving, testing, MLOps
- Data & applications: Python, SQL, FastAPI, PostgreSQL, vector databases, Streamlit
- Cloud: AWS, GCP, Azure
- Built agentic, Arabic legal-AI workflows for research, guided data collection, and document generation.
- Migrated ML serving infrastructure to Ray Serve on Kubernetes for autoscaling and fractional GPU/CPU allocation.
- Improved Arabic manuscript text extraction through OCR and autocorrection systems.