Zero-shot to few-shot, chain-of-thought, and structured output patterns.
Move from "prompts that work sometimes" to prompts you can put in production — structured, tested, versioned, and cheap enough to run at scale.
Anyone whose product depends on an LLM output being correct: engineers wiring model calls, PMs writing spec, and analysts using models to think faster.
The guides are numbered — read in order for the curriculum path, or jump straight to the one you need. Each card is self-contained.
Why prompts shift outputs, tied back to next-token prediction.
The universal prompt skeleton.
Showing vs telling; when examples beat instructions.
Forcing intermediate steps to improve accuracy.
The consolidated pattern toolkit.
Linear vs branching reasoning and when each wins.
Schemas, JSON-only, validation and retries.
Separating instructions from data; first security awareness.
Setting durable behaviour and voice.
Grounding, say-I-don't-know, uncertainty handling.
Breaking big tasks into linked prompts.
Test cases, comparison, avoiding tuning by vibes.
Versioning, testing, treating prompts as deployable artifacts.