AI Architect. Sovereign, model-agnostic LLM, RAG and agentic AI architecture for regulated banking.
Look. In a bank, a clean demo proves NOTHING. The system is ready when it holds up under production load and adversarial testing. Not before.
That's the whole job. Governance is architecture. Observability, audit trails and human oversight are built in from day one, not bolted on later.
- Sovereign AI platforms for banks: data residency, identity and audit in the design, so bank data stays in the country.
- Model-agnostic LLM gateways: swap frontier and open-source models without rewriting application code.
- RAG and agentic workflows that survive production: durable, auditable, human-in-the-loop.
- LLM red teaming for banking: defensive testing against finance-specific risks before go-live and after every model update.
18+ years of software architecture across banking, aviation, security and embedded systems. My background is low-level: systems work and reverse engineering. Same precision, now on AI platforms.
- llm-redteam-banking-checklist: an open-source go-live gate for LLM red teaming in banking. Nine harm categories (seven mapped to the OWASP LLM Top 10 2025), five go/no-go gates, a markdown evidence pack. Runs in pytest and GitHub Actions. A red gate FAILS the build. Apache-2.0.
- LLM Red Teaming for Banking Severity Scorer (Hugging Face Space): the same scorer in the browser. No attack prompts, no model calls, nothing you score is uploaded.
- LLM Red Teaming for Banking: A UAE Go-Live Checklist
- Architecting a Sovereign GenAI Platform for Regulated Banking in the UAE
- DIFC Regulation 10: The AI Architecture It Demands
- More on Medium and dev.to
ercin.info · AI architect in Dubai: what I build · X @aiarchitect · Hugging Face · Medium · dev.to · Stack Overflow