Talking to a model is easy. Directing one with precision — reliably, repeatably, in production — is a craft. This module teaches you to speak to LLMs the way professionals do, and to build against real APIs from day one.
Anyone can type a question into a chatbot. An engineer designs prompts that behave the same way on the thousandth run as on the first, returns data a program can parse, and knows exactly what each API call costs.
You will write real code against the OpenAI and Anthropic Claude APIs, in a room with mentors, from your very first session in this module.
The core patterns behind reliable prompts: role and system prompts, delimiters, instruction ordering, and how to iterate on a prompt like you would on code.
Getting models to reason step by step, and teaching them by example — when few-shot prompting helps, when it hurts, and how to measure the difference.
Turning free-form text into data your programs can trust: JSON mode, schemas, validation, and recovery strategies when a model goes off-script.
The bridge to agents: letting a model call your functions, defining tools well, and handling the full request-response loop safely.
Hands-on work with both major APIs — messages, streaming, system prompts and model selection — so you are fluent on either side of the ecosystem.
What professionals watch in production: token budgeting, caching, retries, timeouts and graceful degradation when an API has a bad day.
Real APIs, real code, real engineering habits — taught free, offline, alongside mentors who ship AI for a living.