At Mozn, I developed a multilingual, multi-task embedding model for enterprise retrieval that outperformed Qwen3-Embedding and BGE-M3 across most internal and public benchmarks. I also built synthetic-data pipelines and extended embedding-serving infrastructure for multiple models.
At Zinki.AI, I developed production document-understanding and Arabic OCR models, and built the end-to-end semantic search and RAG pipeline for SEBR. Independently, I co-developed Makeen for Arabic function calling and developed SanadAI for Islamic citation verification; both were presented in accepted ArabicNLP 2026 system-description papers.

