Prompt Templates

Introduction

A prompt template is a parameterised string with placeholders for dynamic values. Templates turn prompts into reusable, versioned, testable software components.

Beginner analogy: Like email templates with merge fields. Same letter, different recipient, predictable behaviour.

Understanding the topic

Core concepts to understand:

• Use placeholders: {user_name}, {ticket_text}, {context}.
• Store templates in files (md, jinja, .prompt) — not inline strings.
• Version like code — semver or git-based.
• Render with a templating engine (Jinja, Handlebars, Python f-strings).
• Tools: LangChain PromptTemplates, LlamaIndex, Promptfoo.

Syntax reference

Visual workflow / architecture:

template.md:

## SYSTEM
You are a {role}.

## USER
Summarise this ticket:
{ticket_text}

## FORMAT
Return JSON: { "summary": ..., "urgency": ... }

Real-world use

Every serious AI product stores prompts as templates with version control. Inline string prompts in business code are an anti-pattern.

Best practices

• Templates live in /prompts directory, NOT in business logic.
• Each template has tests + evals.
• Track template version in API logs.

Hands-on exercise

Interview preparation — practice these questions:

• Q1. Why use prompt templates?
• Q2. Where should templates live in a codebase?
• Q3. How do you version prompts?
• Q4. Tools for template management?