Summary
Written in English from the original postingThe role involves deploying and operating an LLM model-serving platform using vLLM on on-premise GPU infrastructure. It includes building and running a LiteLLM routing layer, tuning GPU serving, and wiring multi-environment inference across private cloud, AWS, and Azure. The engineer will also own observability, fallback, cost, and performance of the serving layer.
Job description
SDG Group is a global enterprise data, analytics and AI consultancy. We design and build data and AI solutions for our clients. You will stand up and run the model-serving platform behind a production Agentic AI build: open-source models self-hosted on-premise with vLLM, a LiteLLM router in front, and a model garden spanning private cloud, a regional provider and a hyperscaler region, with routing driven by cost, performance and data sensitivity.
What you will do
Deploy and operate self-hosted open-source models on vLLM on-premise GPU infrastructure.
Build and run the LiteLLM routing layer and the model-garden register of approved models and the data each may receive.
Size and tune GPU serving (quantization, KV-cache and concurrency planning) so the on-premise footprint holds at load.
Wire multi-environment inference across private cloud, a regional provider and AWS Bedrock or Azure AI Foundry, with routing rules enforcing data residency.
Own observability, fallback, cost and performance of the serving layer.
What you bring
Solid MLOps or ML platform engineering, with real LLM serving experience (vLLM, TGI or equivalent).
A model gateway or router in production (LiteLLM or similar).
GPU infrastructure depth: CUDA, VRAM and KV-cache sizing, quantization trade-offs, and concurrency.
Kubernetes and containers, CI/CD, and cloud (AWS and/or Azure); monitoring and cost control.
Nice to have
Hands-on with open-weight models such as Qwen, GPT-OSS and Kimi.
Sovereign or data-residency-constrained deployments; Terraform or other infrastructure as code; telecom scale.
Why SDG Group
Own the serving and routing backbone of a large enterprise Agentic AI platform, including real on-premise GPU and multi-cloud model infrastructure. We welcome strong candidates regardless of current location, provided they can be based in Cairo and travel as the engagement requires.
The weekly Digest
The week's most transparent data engineering jobs in your inbox once a week.
Once a week, a few new jobs that state their salary, location and stack, plus links to the rest. No open tracking; clicks are counted per issue, not per person. Unsubscribe with one click.
