Production prompting from an NVIDIA and open-model angle: system prompts for OpenAI-compatible endpoints (NIM, vLLM), sampling parameters (temperature, top-p, top-k, seed), chat templates for open-weight models, strict structured outputs, tool use and MCP, NeMo guardrails, security (injection, PII, EU AI Act) and productionization (versioning, evaluation, observability, costs). 19 lessons, a capstone and a graded final project.
Beyond the single prompt: chaining, prompt-as-program, tool use, and adaptive thinking. This is where you stop writing prompts and start designing systems.