Skip to content

Level 2 · Intermediate Techniques

Goal: build a toolbox of techniques that work across models and tasks, and learn when each one is — and isn't — worth using.

Level 1 was about writing one good prompt. Level 2 is about the patterns that show up again and again in real work: getting a model to reason before answering, splitting big jobs into chains of small prompts, producing machine-readable output, writing system prompts, controlling voice and length precisely, and the everyday workhorse tasks of extraction, classification, summarization, and transformation. The level ends with a project that combines several techniques into a small pipeline.

From here on, lessons include short Python examples. They use only the standard library and a fake_model() stand-in so they run without an API key; swap in a real model call when you're ready. How to call specific provider APIs is covered in the LLM Dev Mastery Path — this course stays focused on what goes into the call.

Modules

  1. Reasoning Prompts & Chain-of-Thought — asking for reasoning, when it helps, when it doesn't, and reasoning models
  2. Decomposition & Prompt Chaining — splitting a task into steps whose outputs feed each other
  3. Structured Output: JSON & Schemas — reliable machine-readable output, validation, and repair
  4. System Prompts — what belongs in a system prompt versus the user turn, and how to organize one
  5. Controlling Tone, Length & Style — style specifications, voice samples, and reading-level targets
  6. Extraction & Classification Prompts — pulling fields out of text and assigning labels consistently
  7. Summarization Prompts — purpose-driven summaries, faithfulness, and long documents
  8. Prompt Templates & Variables — turning prompts into safe, reusable templates
  9. Transformation Prompts: Rewrite, Translate, Convert — changing form without changing meaning
  10. Project — A Support-Ticket Triage Chain — classify, extract, and draft replies in a tested three-step chain

Before you start

  • Complete Level 1, or be comfortable with its ideas: context, delimiting, few-shot examples, format templates, and iterating against a small test set.
  • Python 3.9 or newer for the optional code examples.