Sovereign AI architecture for regulated banks in the UAE.
AI Architect. I own end-to-end RAG and agentic architecture on sovereign cloud infrastructure: model-agnostic, audited, and running in production for core banking.
Request flow through the sovereign AI platform: Client → model-agnostic gateway → vector retrieval → agentic orchestration → knowledge graph → identity and access → Response.
Don't take my word for it. Check it.
Look. Anybody can write "banking AI expert" on a website. Here is what you can verify yourself, today.
- A citable checklist with a DOI. The LLM red-teaming checklist for banking is archived on Zenodo: 10.5281/zenodo.23536860, released 21 Oct 2026, source on GitHub.
- A working scorer, not a slide. The LLM Red Teaming for Banking Severity Scorer runs on Hugging Face. Nine harm categories (seven mapped to the OWASP LLM Top 10 2025).
- The article Google puts first. LLM red teaming for banking in the UAE ranked #1 on Google UAE for "LLM red teaming for banking" on 23 Oct 2026.
- A named failure mode. Borrowed Clearance: when a bank's AI agent moves money based on a check it didn't run and can't verify. Coined by Ercin Dedeoglu, 21 Oct 2026.
What I do
I design the architecture that lets enterprises run generative AI in environments where mistakes are not an option. I do this as an AI architect in Dubai, for banks across the UAE, where data residency and DIFC/CBUAE expectations shape the design. That means provider-agnostic LLM gateways, retrieval and agentic patterns that survive production load, and governance — observability, audit trails, human-in-the-loop — treated as first-class, not afterthoughts. My background is low-level: reverse engineering and systems work. I bring that precision to AI platforms.
Selected work
Sovereign RAG for core banking
A production retrieval system unifying structured and unstructured data inside a regulated banking suite.
Read →Model-agnostic LLM gateway
Provider-independent routing so the platform swaps models without touching application code.
Read →Agentic workflows, safely
Durable, auditable agent execution with guardrails against the failure modes that kill enterprise pilots.
Read →