02 · Prompting & Control

How you talk to the model. High ROI — tackle this early. Good prompting technique pays off across every other section.

Step	Topic	One-liner	Status
1	System prompt	Setting the model's persona, constraints, and context before the conversation starts	🟢
2	Prompt engineering	The craft of writing instructions the model follows reliably	🟢
3	Few-shot & zero-shot	Guiding the model with examples vs. instructions alone	🟢
4	In-context learning (ICL)	How models adapt behavior from examples in the prompt window	🔴
5	Temperature, Top-p & sampling	The parameters that control randomness and diversity in generation	🟢
6	Structured output & JSON mode	Forcing the model to emit valid JSON, XML, or schema-conforming output	🟢
7	Prompt versioning	Managing prompts like code — version control, rollback, A/B testing	🟢
8	Constitutional AI	Self-critique and revision loops to align model output with principles	🔴
9	System card	The documentation artifact that describes a model's intended use and limitations	🔴
10	Context Management Patterns	Handle long conversations, sliding windows, and token budgets in production.	🔴

---

← Previous section: 01 · Model Inference Core | Next section → 03 · Serving Infrastructure