Omar Elsherif

I'm Omar, an AI Engineer who builds LLM systems that hold up in production: agents, RAG pipelines, and the evaluation and serving infrastructure behind them.

At Penta-b, I build bilingual (Arabic/English) LLM agents that drive an interactive GIS map through natural-language tool-calling. I own the agent's evaluation harness (multi-turn, LLM-as-judge, gated in CI with Langfuse tracing). I cut input tokens by ~48% through prompt caching and per-turn tool selection, which projects to about $140k/month in savings at scale. I also lead our move to on-prem open-weight models, benchmarking 27B Qwen models at roughly 9× lower cost and building the vLLM serving path.

Before that, at GRINDA AI, I spent two years building multilingual RAG systems and agents across 15+ languages. I fine-tuned and aligned Qwen models with LoRA, DPO, and RLAIF, and optimized multi-GPU training and inference on H100 clusters with DeepSpeed and vLLM.

🎓 Always learning

• Master of Science in AI Engineering (MSAIE), Quantic School of Business and Technology
• Islamic Finance studies at the Islamic Finance Institute, preparing for AAOIFI certification
• B.Sc. Medical Informatics (CS), Benha University, GPA 3.89/4.0, Ideal Student of the Year 2023

🚀 Where I'm heading
Long term, I want to build businesses, not just systems. I'm especially interested in where AI engineering meets real markets, including ethical and Shariah-compliant finance.