Cairo, Egypt
I ship LLM systems that work in production. Not prototypes. Not demos. My focus is Arabic-first AI: building retrieval systems, agents, and chatbots that handle MSA and dialects at scale. I've automated 40-60% of customer queries on WhatsApp, built AI assistants for Saudi ministries, and led on-prem GenAI deployments that reduced manual work by 25-35%. I care about the hard parts: evaluation that catches real failures, inference costs that don't explode, and retrieval that actually retrieves. I've fine-tuned LLMs on domain data, optimized RAG pipelines, and built observability into LLM systems from day one. Core stack: Python, PyTorch, vLLM, llama.cpp, LangChain, Kubernetes, AWS. Published researcher in Arabic NLP. Mentor at lablab.ai. Currently leading AI at ECC and consulting for legal AI at i-Legal. Let's talk if you're building Arabic AI or deploying LLMs in production.
• Advised founders and product leadership on comprehensive AI strategies, focusing on LLM selection and deployment patterns for legal workloads. • Designed and reviewed architecture for bilingual legal NLP components, enhancing contract understanding and clause retrieval. • Defined experimentation frameworks and QA guidelines to ensure model quality and data privacy in high-stakes legal contexts.
• Spearheaded AI transformation across sales, operations, HR, and finance, achieving a 25-35% reduction in manual processing time. • Developed on-prem AI tools and chatbots, decreasing internal request turnaround time by approximately 40%. • Automated 50-70% of recurring employee queries, allowing staff to concentrate on higher-value tasks. • Led on-prem GenAI and LLM assistants for internal stakeholders, using RAG over ERP/CRM data and secure Arabic LLMs.
• Developed and optimized Arabic-first conversational AI chatbots for multiple platforms including WhatsApp and Facebook Messenger. • Enhanced intent classification and entity extraction for MSA and dialectal Arabic, boosting understanding accuracy by 8-15%. • Collaborated with clients across MENA to successfully launch over 10 production chatbots, significantly reducing human support workload.