Module Overview
Deep dive into prompting patterns and design techniques that consistently elicit desired outputs: zero/one/few-shot patterns, templates, output schemas, persona/role-play, and multi-step prompt chaining. Emphasis on reusability and testing.
Learning Objectives
- Choose and apply the optimal shot pattern for a given task and dataset size.
- Design reusable, parameterized prompt templates that enforce output schemas (JSON, CSV, bullet lists).
- Implement prompt chaining patterns to break complex tasks into manageable steps.
- Build a small test harness to evaluate template stability across inputs.
Lesson-by-Lesson Breakdown
Shot patterns: tradeoffs and performance examples (zero/one/few).
Template engineering: parameterization, placeholders, and environmental variables.
Output shaping: schema enforcement, delimiters, and parsing reliability.
Role-play & persona control: tone, instruction bias, and consistent voice.
Chaining & decomposition: design patterns for multi-step generation.
Building a template library and tagging for reuse.
Peer testing: structured A/B checks on templates.
Hands-on Activities & Deliverables
Activities
Create a template library (minimum: summary template, Q&A template, lesson-plan generator) with test cases and expected schema.
📦 Deliverable
Template repo + README describing usage, parameters, sample inputs & expected outputs.
Required Tools & Readings
Example template repositories, simple script for running prompts at scale (CSV in/out), documentation on model parameters.
Assessment & Rubric
- Reusability & clarity of templates35%
- Schema adherence30%
- Testing coverage and results20%
- Documentation quality15%
Prerequisites
Completion of Module 1 or equivalent understanding of LLM basics.
Parent-Friendly Value
Produces repeatable prompt 'tools' students can reuse across tasks (homework help, lesson creation, content summarization).