Mostafa E.

I'm an AI/ML engineer specializing in Arabic NLP, speech AI, and production RAG systems. I build and ship real systems, not just notebooks. Recent work: Fine-tuned XLM-RoBERTa-large for Arabic punctuation restoration (Naqta), published on HuggingFace, surpassing IEEE 2024 SOTA on the benchmark task. Fine-tuned wav2vec2-large-xlsr-53 for Arabic phoneme-level ASR, 3.95% PER on held-out test data, also published on HuggingFace. Built and run a production voice AI pipeline (ASR, LLM, TTS) at my current role, served on AWS with autoscaling and batched inference, including domain-based model swapping so the system routes to different fine-tuned models depending on context, without downtime. Built a full-stack RAG platform for legal documents on Qdrant, with LangChain/LangGraph handling the agent layer, Redis and Supabase on the backend, and a React/Next.js frontend. A chunking strategy I developed improved retrieval precision by 35%, with sub-500ms latency. What I bring to a project: Real production experience, not just research or prototypes. I deploy with Docker and Kubernetes, and I validate AI output rather than trusting it blindly, every system I build treats model output as something to check, not something to assume is correct. Strong communication in both Arabic and English (native/bilingual in both), so I work well with clients across MENA and international markets. I'm comfortable owning a project end to end: requirements, architecture, implementation, deployment, and documentation. Happy to share links to published models and project details on request.