| Part 1: Prompting, Context and Grounding |
How an LLM produces text; choosing and running an open-weight model; prompting and structured outputs; sampling; context engineering; grounding answers with retrieval; and the limits of all of these |
Open |
| Part 2: Prompt Engineering, Memory and Control |
Good and bad prompts; roles, system messages and chat templates; zero-shot, few-shot, chain-of-thought, meta-prompting and self-consistency; short-term, semantic and episodic memory and the context window; temperature, top-k and top-p; prompt injection and defenses |
Open |
| Lab 1: Prompting, Measured |
Test each prompting technique on Arabic tweets labeled for sentiment, sarcasm and dialect (ArSarcasm-v2): vague against specific, zero-shot against few-shot, structured output with validation and retry, and a prompt injection measured with and without a defense. |
Notebook Colab |
