Building LLM Applications: Architecture to Production

For software engineers shipping LLM features in real products: how to pick between workflows and agents, get typed outputs and reliable tool calls, build evals you can trust, control cost and latency, and handle failures and prompt injection in production.

5 sections ยท 10 lessons

Course outline

Choosing the Architecture

  1. Workflows Before Agents
  2. Five Workflow Patterns
  3. Choosing an Architecture for Email Triage

Structured Outputs and Tools

  1. Structured Outputs That Hold
  2. Designing Tools the Model Can Use

Evals You Can Trust

  1. Evals Built from Real Failures
  2. Graders and Judges That Hold Up
  3. Building a First Eval Suite

Cost and Latency

  1. Where the Token Bill Comes From
  2. Making Responses Feel Fast

Failure and Security

  1. When Calls and Answers Fail
  2. Prompt Injection and Write Actions
  3. Launch Review for an Inbox Assistant

Start learning with Wondering