Build Your AI Infrastructure Foundation. Learn what GPUs, model servers, and LLM routing systems actually do — no machine learning background needed.

Every AI product you use runs on infrastructure someone had to build. This course shows you what that infrastructure looks like, piece by piece, so you can reason about it instead of guessing. You'll follow a single path from start to finish: serving a large language model to many users at the same time. It begins with a question typed into a chat box and works down to the hardware that answers it. Each lesson introduces a real problem — the model won't fit in GPU memory, the first answer takes too long to start, or the load balancer sends a request to the wrong server — then walks through the solution and the new challenge it creates. Across four modules you'll build a clear mental model of the full stack: what a model actually is as a file of numbers, why GPUs run inference instead of CPUs, what vLLM does as a model server, how tokens are generated one at a time, why the KV cache is the biggest performance lever in LLM serving, how batching and tensor parallelism scale to many users and large models, and why traditional load balancing fails for this traffic. The course closes with llm-d, the open-source project from Red Hat, Google, IBM, and NVIDIA that uses cache-aware routing and prefill/decode disaggregation on Kubernetes. A free hands-on lab lets you build a working vLLM server that handles concurrent requests. Role-play exercises help you practice explaining these concepts to stakeholders. Who this is for: DevOps engineers, cloud architects, and sysadmins ready to add AI workloads to their infrastructure skill set.