Building Production AI Systems

Most AI projects stall between a working demo and a system an organization can actually deploy. This course closes that gap by teaching AI development the way production teams approach it: architecture first, with prompting treated as an implementation detail rather than the starting point. Over six weeks, students build a single enterprise AI application, adding one production capability each week. They begin by designing the system architecture and understanding where inference, retrieval, tools, and memory belong. They then implement context engineering with retrieval, grounding, citations, and structured outputs; orchestrate deterministic workflows with routing, state management, retries, and human approval steps; and introduce agents only where the architecture justifies them, with equal attention to when agents are the wrong choice. Throughout, students grow an evaluation suite alongside the product, and in the final weeks harden the system with observability, tracing, cost and latency optimization, prompt versioning, security testing, and deployment. Students finish with one deployed, instrumented, evaluated application suitable for a professional portfolio rather than a series of disconnected exercises.