Chapter 12: Prompt Engineering & In-Context Learning¶

Design inputs that get reliable, useful behavior from LLMs—prompt anatomy, zero/few-shot, chain-of-thought, ReAct, structured outputs, evaluation, injection defenses, and a versioned prompt registry.

Metadata¶

Learning Objectives¶

• Decompose a prompt into instruction, context, input, and output spec
• Apply zero-shot, few-shot, and in-context learning patterns
• Use chain-of-thought, self-consistency, ReAct, and tool/function calling
• Produce structured outputs with Pydantic schemas and safe parsers
• Evaluate prompts with golden datasets, graders, and A/B tests with CIs
• Defend against prompt injection and ship versioned prompts to production

What's Included¶

Notebooks¶

Scripts¶

• config.py — Chapter config, mock-LLM toggle, registry paths
• prompt_templates.py — Reusable Jinja-style templates for zero-shot, few-shot, CoT, ReAct
• llm_clients.py — BaseLLMClient, MockLLMClient, optional adapter for OpenAI / Anthropic
• evaluation_utils.py — Golden datasets, graders, A/B tester with bootstrap CIs

Exercises¶

• Problem Set 1 (notebook) — Rewrite a vague prompt, build few-shot examples, structured-output schema, classify a tricky example, count tokens, parse JSON
• Problem Set 2 (notebook) — Self-consistency, eval harness, injection detection, A/B test, ReAct loop, versioned registry
• Solutions — In exercises/solutions/ (notebooks and solutions.py for CI)

Diagrams (Mermaid)¶

• prompt_anatomy.mermaid, chain_of_thought.mermaid, evaluation_loop.mermaid

Read Online¶

• 12.1 Introduction — Prompt anatomy, zero/few-shot, in-context learning, structured outputs
• 12.2 Intermediate — Chain-of-thought, self-consistency, ReAct, tool/function calling
• 12.3 Advanced — Evaluation, A/B tests, injection defenses, versioning, production

Or try the code in the Playground.

How to Use This Chapter¶

1. Clone and install dependencies

git clone https://github.com/luigipascal/berta-chapters.git
cd berta-chapters
pip install -r requirements.txt

2. Navigate to the chapter

cd chapters/chapter-12-prompt-engineering-and-in-context-learning
pip install -r requirements.txt

3. (Optional) Wire up a real provider

pip install openai anthropic
# All notebooks default to the bundled MockLLMClient — no API keys required.

4. Launch Jupyter

jupyter notebook notebooks/01_prompt_basics.ipynb

Created by Luigi Pascal Rondanini | Generated by Berta AI