Abu Dhabi, UAE · Taking 1–2 side engagements per quarter
Forward deployed AI engineer · Fintech & finance ops
I ship AI systems that survive production.
I build LLM and agent systems with the evals, guardrails and audit trail that regulated businesses need, so they hold up with real customers and real money. 10+ years of production backend and fintech engineering behind every build.
30 →to 2 minStore checkout wait after self-checkout launch
30 →to 10 minDeploy time after Kubernetes migration
UAEFintech integrations with central-bank & government systems
/ 01 — Forward deployed
Most AI pilots stall between the demo and production. That gap is where I work.
Pilots stall on the parts a demo never touches: messy real data, edge cases nobody wrote down, compliance questions, and no one owning the system after launch. Working forward deployed, inside your team rather than from a slide deck, is how those surface early instead of in production.
/ 02 — Engagements
Three ways to work together.
012 weeks · fixed scope
AI Production Readiness Review
Find out exactly what stands between your AI pilot and production.
What you get
Architecture & data-flow review of your LLM / agent system
Eval plan: what to measure, golden datasets, pass/fail thresholds
Risk review mapped to the OWASP Top 10 for LLM Applications
Cost, latency and observability assessment
Prioritised roadmap your team can execute — walked through live
Ideal if You have a working demo or pilot and leadership is asking "can we put this in front of customers?"
How I architect AI for workflows where a wrong answer costs money: deterministic engines generate candidates, the model only chooses among them under a strict schema, and every decision is traceable.
0Model calls in the deterministic-only baseline
≥85%Auto-match target by volume before MVP (exit gate, not a result)
A high-availability messaging layer in Go and Java that filtered, processed and logged customer messages at scale — plus the Kubernetes migration that cut deploy time by two-thirds.
2,000 TPSSustained messaging throughput
30 →to 10 minDeploy time after moving from VM-based Docker to Kubernetes
Designed and built a self-checkout WeChat mini-program for retail stores, integrating product, payment and store systems so customers could skip the queue.
My approach is deliberately practical: understand the constraints, make the hard decisions visible, and build for the operating reality—not just the happy path.
01
Start with the operating model
Clarify the customer, workflow, constraints, and failure modes before selecting tools.
02
Design clear boundaries
Use focused domains, explicit contracts, and simple interfaces so teams can move independently.
03
Make it measurable
Evals, traces and a decision log from day one — so you can prove the system works, not just believe it.
04
Ship a path forward
Leave teams with an architecture and implementation plan they can confidently extend.
Tell me the process, the systems it touches and what “done” looks like. I reply within two working days, and I’ll tell you honestly if I’m not the right fit.