AI Engineer & Full-Stack Developer — Egypt

I build AI systems that are meant to survive contact with real users: guardrailed, tested, documented, and deployed.

Most AI demos break the moment they meet production. A model that answers well in a notebook still has to be constrained, validated, monitored, and made safe to run against real data.

That gap is my work. I treat model output as untrusted input, put an explicit boundary in front of it, and make the system prove itself with tests before it ships.

• Guardrailed text-to-SQL — AST-level SQL validation instead of keyword filters
• Graph RAG and knowledge-graph retrieval over messy, real-world records
• Bilingual (Arabic / English) NLP pipelines
• Agentic workflows with bounded, auditable tool use

Natural-language analytics over a SQL database, built so that model-generated SQL can never be trusted blindly.

Generated SQL is parsed into an AST and checked by policy — single read-only statement, schema allowlist, no mutations, no data-modifying CTEs, no system catalogs, enforced row limits — then executed through a read-only connection with a fail-closed authorizer. Ships with a deterministic offline provider so the whole pipeline runs, and is tested, without any API key.

Python Gemini SQLGlot Streamlit Docker pytest · 184 offline tests · CI green · MIT

Bilingual, human-in-the-loop lost-item matching for airport operations — explainable scoring, Graph RAG, and image matching, with a complete local demo mode.

Built as a full production-shaped system rather than a notebook: API, database, cache, typed frontend, containers, and infrastructure definitions.

Python FastAPI PostgreSQL Redis Graph RAG Computer Vision React TypeScript Docker Azure · MIT

Big Marvel fan — less for the capes, more for the workshop scenes. The appeal was never the suit; it was someone building, testing, and iterating until the thing actually held up under load.

That's the same instinct I bring to engineering: build it, then try hard to break it before anyone else does.

Outside that: chess, football, and reading about how large systems fail.

Email — abdallah.hosnicv@gmail.com  · 

Open to AI engineering roles and freelance work on production LLM, RAG, and computer-vision systems.