Firas Al Wadhahi

I lead AI/ML infrastructure at Rihal, a technology company headquartered in Muscat, Oman. My work centres on building and operating production-grade systems for large language models — from bare-metal GPU provisioning through inference optimisation to the agentic products that ship to end users. The mandate is clear: get research-grade models into production reliably, at the latency and cost that actually matter.

Our compute platform runs on NVIDIA DGX H200 hardware. I own the full stack: SLURM scheduling, Kubernetes orchestration, vLLM serving, and the platform abstraction layers that let product teams deploy GenAI capabilities without needing to reason about what sits beneath. On-premise by design — data sovereignty and inference cost are non-negotiable constraints in the markets we serve.

Before ML infrastructure I worked across data engineering and applied research. I care about systems that are honest about their limits — the hard constraints of hardware, latency, and accuracy matter more to me than benchmark leaderboards. If a model cannot serve a request in under two seconds on the worst-case hardware, the capability does not exist.