Genai Patterns Prompting

# GenAI Patterns & Prompting

## Why it matters
- Pattern-driven design turns "prompt tinkering" into **repeatable systems** with measurable quality.
- Good prompting + retrieval + evaluation unlocks **reliable outputs** and lower cost/latency.
- Shared patterns enable **governance**, faster onboarding, and vendor portability.

## Core concepts
- **Inputs & controls**: instructions, few-shot examples, tools/functions, system rules, temperature/top-p, stop sequences.  
- **Context window economics**: tokens = money + latency; retrieve only what's needed.  
- **Tool use**: the model calls functions (search, calculators, CRUD) to ground and act.  
- **Retrieval-augmented generation (RAG)**: fetch source facts → generate grounded answers.  
- **Agents**: multi-step planners using tools + memory; powerful, but require guardrails.  
- **Evaluation**: faithfulness/groundedness, task success, cost, latency, safety.

## Diagram — choosing a GenAI pattern