Back to Articles/AI System Architecture Concepts

Sections

Introduction 1. Vector Database 2. Embedding Model 3. RAG (Retrieval Augmented Generation)Advanced RAG Agentic RAG Modular RAG 4. Prompt Engineering 5. Semantic Cache 6. MCP (Model Context Protocol)7. AI Agent 8. Function Calling 9. Agent Guardrails

4. Prompt Engineering

6 min read

The art and science of crafting inputs (prompts) that guide LLMs to understand the task, context, and constraints — producing better, more reliable outputs.

Prompt Engineering

What is Prompt Engineering?
Anatomy of a Great Prompt
Core Prompting Techniques
Advanced Techniques
System Prompts
Common Mistakes to Avoid
Prompt Templates
Key Takeaway

What is Prompt Engineering?

Prompt engineering is the practice of designing and refining the instructions you give to a Large Language Model to elicit the best possible output. It is part craft, part science — a well-engineered prompt can be the difference between a generic, inaccurate response and a precise, expert-level answer.

Poor prompt:   "Summarise this."
               → Vague, unpredictable length and style

Good prompt:   "You are a senior technical writer. Summarise the following 
                document in exactly 3 bullet points. Each bullet should be 
                under 20 words and highlight a key benefit for developers.
                Document: {document}"
               → Consistent, targeted, useful output

Anatomy of a Great Prompt

Every well-structured prompt contains some or all of these components:

┌──────────────────────────────────────────────────────────┐
│  ROLE / PERSONA                                          │
│  "You are an expert Python developer..."                 │
├──────────────────────────────────────────────────────────┤
│  TASK / INSTRUCTION                                      │
│  "Review the following code and identify bugs..."        │
├──────────────────────────────────────────────────────────┤
│  CONTEXT                                                 │
│  "This is a FastAPI service handling payment webhooks."  │
├──────────────────────────────────────────────────────────┤
│  INPUT / DATA                                            │
│  "Code: {code_snippet}"                                  │
├──────────────────────────────────────────────────────────┤
│  EXAMPLES (few-shot)                                     │
│  "Example bug: Missing null check → Example fix: ..."   │
├──────────────────────────────────────────────────────────┤
│  OUTPUT FORMAT                                           │
│  "Return a JSON array: [{line, severity, description}]"  │
└──────────────────────────────────────────────────────────┘

Core Prompting Techniques

1. Zero-Shot Prompting

Ask the model to perform a task with no examples. Works for simple, well-known tasks.

Classify the sentiment of this review: "The product broke after 2 days."
Answer: Negative

2. Few-Shot Prompting

Provide examples to show the model the expected input→output pattern. Dramatically improves accuracy for novel or nuanced tasks.

Classify sentiment:

Review: "Absolutely love it!" → Positive
Review: "It's okay, nothing special." → Neutral
Review: "Completely useless, waste of money." → Negative

Review: "Arrived on time, works as expected." →

3. Chain-of-Thought (CoT) Prompting

Instruct the model to reason step-by-step before giving the final answer. Significantly improves performance on reasoning and maths tasks.

Q: A store has 120 items. 30% are electronics. 
   If 25% of electronics are on sale, how many electronics are on sale?
   
Think step by step:
1. Electronics = 30% × 120 = 36 items
2. On sale = 25% × 36 = 9 items
Answer: 9 electronics are on sale.

4. Role / Persona Prompting

Assign a role to the model to activate domain expertise and appropriate tone.

You are a senior DevSecOps engineer with 15 years of experience.
Review this Dockerfile for security vulnerabilities and explain each risk in plain English.

5. Self-Consistency

Generate multiple reasoning chains with temperature > 0, then take a majority vote on the final answer. Improves reliability on complex tasks.

Advanced Techniques

ReAct (Reasoning + Acting)

Interleave reasoning steps with tool calls. The model thinks, then acts, then observes.

Thought: I need to find the current weather in London.
Action: search("current weather London")
Observation: 18°C, partly cloudy
Thought: I now have the weather. I can answer.
Answer: The weather in London is currently 18°C and partly cloudy.

Tree of Thoughts (ToT)

Explore multiple reasoning paths simultaneously and evaluate which branch is most promising — like a search tree over thought space.

Prompt Chaining

Break complex tasks into a sequence of simpler prompts, where the output of one becomes the input of the next.

Prompt 1: Extract key facts from this document → {facts}
Prompt 2: Based on these facts: {facts}, write an executive summary
Prompt 3: Translate the summary into formal French

Meta-Prompting

Ask the LLM to generate or improve the prompt itself.

I need to write a prompt that will get an LLM to generate unit tests for Python code.
The prompt should cover: test naming conventions, edge cases, mocking dependencies.
Write an optimal prompt for this task.

System Prompts

The system prompt sets persistent context, persona, and rules for the entire conversation. It is separate from the user message and given higher authority.

messages = [
    {
        "role": "system",
        "content": """You are a helpful customer support agent for Acme Corp.
        Rules:
        - Only answer questions about Acme Corp products.
        - If asked about competitors, politely decline.
        - Always be friendly and concise.
        - If you don't know something, say so — never guess.
        - Format lists as bullet points."""
    },
    {
        "role": "user", 
        "content": "What's your return policy?"
    }
]

Best practices for system prompts:

Put the most important constraints first
Use numbered lists for ordered rules
Be explicit — don't assume the model will infer intent
Test edge cases to ensure rules hold

Common Mistakes to Avoid

❌ Mistake	✅ Fix
Vague instructions: "Write something good"	Specific: "Write a 200-word product description for a wireless keyboard, highlighting battery life and quiet keys"
No output format specified	Add: "Return as JSON: {title, description, price}"
Overloading one prompt	Break into a chain of focused prompts
Ignoring model's training cutoff	State the date and provide fresh context
No persona/tone direction	Specify: "Write in a professional but friendly tone"
Prompt injection risk	Sanitise user input before inserting into prompt templates

Prompt Templates

Classification Template

You are an expert classifier. Classify the following {input_type} into one of these categories: {categories}.

Rules:
- Return only the category name, nothing else.
- If no category fits, return "Other".

{input_type}: {input}
Category:

Summarisation Template

You are a professional summariser. Summarise the following text in {format}.

Requirements:
- Length: {length}
- Audience: {audience}
- Tone: {tone}

Text:
{text}

Summary:

Code Review Template

You are a senior {language} developer. Review the following code for:
1. Bugs and potential errors
2. Security vulnerabilities
3. Performance issues
4. Code style and readability

For each issue found, provide:
- Line number (if applicable)
- Severity: Critical / Major / Minor
- Description of the issue
- Suggested fix

Code:
```{language}
{code}


---

## Key Takeaway

> **Prompt Engineering** is the skill of communicating effectively with LLMs. Clear roles, structured instructions, relevant context, and well-specified output formats consistently produce better, more reliable results — without any model retraining.

---

**← [RAG](../03-rag/README.md)** | **[Back to Overview](../README.md)** | **[Semantic Cache →](../05-semantic-cache/README.md)**

Table of Contents