Module 19 — Prompt Engineering

Introduction

Prompt Engineering is the practice of designing effective instructions for Large Language Models (LLMs) to improve the quality, accuracy, and reliability of their responses.

A prompt is more than just a question—it defines:

• The task
• The context
• The expected output
• The response format
• The reasoning process
• The constraints

Modern AI systems such as chatbots, coding assistants, document analyzers, and AI agents rely heavily on well-designed prompts.

In this module, you'll learn:

• Zero-shot Prompting
• One-shot Prompting
• Few-shot Prompting
• Chain of Thought (CoT)
• Self Consistency
• Tree of Thoughts (ToT)
• ReAct
• Structured Prompting
• Prompt Templates
• System Prompts
• Build an AI Assistant

How Prompt Engineering Works

A prompt guides the model toward a desired behavior.

textCopy
1                User Prompt
2                     │
3        Instructions + Context
4                     │
5              Large Language Model
6                     │
7            Reasoning & Generation
8                     │
9               Final Response

A well-structured prompt reduces ambiguity and improves response quality.

Anatomy of a Good Prompt

A high-quality prompt usually contains:

textCopy
1Role
2↓
3
4Task
5
6↓
7
8Context
9
10↓
11
12Constraints
13
14↓
15
16Expected Output

Example

textCopy
1Role:
2You are a Python expert.
3
4Task:
5Explain decorators.
6
7Constraints:
8Use beginner-friendly language.
9
10Output:
11Step-by-step explanation with examples.

1. Zero-shot Prompting

What is Zero-shot Prompting?

Zero-shot prompting asks the model to perform a task without providing any examples.

Example

textCopy
1Translate the following sentence into French.
2
3I love machine learning.

The model relies entirely on its pretrained knowledge.

Advantages

• Simple
• Fast
• Minimal prompt size

Limitations

• May produce inconsistent formatting
• Less reliable for specialized tasks

Python Example

pythonCopy
1from transformers import pipeline
2
3generator = pipeline(
4    task="text-generation",
5    model="gpt2"
6)
7
8prompt = "Explain what a Transformer model is."
9
10result = generator(
11    prompt,
12    max_new_tokens=80
13)
14
15print(result[0]["generated_text"])

2. One-shot Prompting

What is One-shot Prompting?

One-shot prompting provides one example before the actual task.

Example

textCopy
1Example
2
3Input:
4Apple
5
6Output:
7Fruit
8
9Now classify:
10
11Carrot

The model learns the desired format from a single demonstration.

Advantages

• Better formatting
• Improved consistency
• Easy to design

3. Few-shot Prompting

What is Few-shot Prompting?

Few-shot prompting provides multiple examples.

Example

textCopy
1Positive → Great product
2
3Negative → Poor quality
4
5Positive → Excellent service
6
7Now classify:
8
9Amazing experience

Advantages

• Better task adaptation
• Improved accuracy
• More consistent outputs

4. Chain of Thought (CoT)

What is Chain of Thought?

Chain of Thought encourages the model to reason step by step before producing a final answer.

Conceptually, this technique decomposes complex problems into intermediate reasoning steps.

Example prompt

textCopy
1Solve the following math problem.
2
3Explain your reasoning step by step before giving the final answer.

Advantages

• Better reasoning
• Improved performance on complex tasks
• More interpretable intermediate logic

Note: When using hosted reasoning models, you generally don't need to ask for internal reasoning. It's often better to ask for a concise explanation of the solution rather than the model's hidden reasoning process.

5. Self Consistency

What is Self Consistency?

Self Consistency improves reasoning by generating multiple candidate solutions and selecting the most consistent final answer.

Workflow

textCopy
1Question
2
3↓
4
5Reasoning Path 1
6
7Reasoning Path 2
8
9Reasoning Path 3
10
11↓
12
13Select Most Consistent Answer

Advantages

• Higher accuracy
• More robust reasoning
• Reduces single-path errors

6. Tree of Thoughts (ToT)

What is Tree of Thoughts?

Tree of Thoughts extends linear reasoning into a tree of alternative solution paths.

Architecture

textCopy
1Problem
2
3├── Thought A
4
5│      ├── A1
6
7│      └── A2
8
9├── Thought B
10
11│      ├── B1
12
13│      └── B2
14
15└── Best Solution

Advantages

• Explores multiple strategies
• Better planning
• Strong performance on search and puzzle tasks

7. ReAct (Reason + Act)

What is ReAct?

ReAct combines reasoning with tool usage.

Instead of answering immediately, the model can decide to use external tools, observe the results, and continue.

Workflow

textCopy
1Question
2
3↓
4
5Reason
6
7↓
8
9Use Tool
10
11↓
12
13Observe
14
15↓
16
17Reason Again
18
19↓
20
21Answer

Applications

• AI agents
• Search systems
• Retrieval-Augmented Generation (RAG)
• Database queries
• API integrations

Example

textCopy
1User:
2What is today's weather in Tokyo?
3
4Reason:
5I need current weather information.
6
7Action:
8Call weather API.
9
10Observation:
1122°C and cloudy.
12
13Final Answer:
14Today's weather in Tokyo is 22°C and cloudy.

8. Structured Prompting

What is Structured Prompting?

Structured prompting organizes instructions into clearly defined sections.

Template

textCopy
1Role:
2...
3
4Objective:
5...
6
7Context:
8...
9
10Constraints:
11...
12
13Output Format:
14...

Advantages

• Better consistency
• Easier debugging
• More predictable outputs

9. Prompt Templates

Prompt templates make prompts reusable by replacing values with variables.

Example

textCopy
1You are a {role}.
2
3Task:
4{task}
5
6Context:
7{context}
8
9Output:
10{format}

Python Example

pythonCopy
1template = """
2You are a {role}.
3
4Task:
5{task}
6
7Context:
8{context}
9
10Output Format:
11{format}
12"""
13
14prompt = template.format(
15    role="Python Expert",
16    task="Explain decorators",
17    context="Audience is beginners",
18    format="Markdown"
19)
20
21print(prompt)

Advantages

• Reusable
• Dynamic
• Easy to automate

10. System Prompts

What are System Prompts?

A system prompt defines the model's overall behavior, personality, and constraints before user messages are processed.

Example

textCopy
1You are an experienced software engineer.
2
3Always provide Python examples.
4
5Use beginner-friendly explanations.
6
7Avoid unnecessary jargon.

System prompts help maintain consistent behavior across an entire conversation.

Prompting Strategy Comparison

Practice 1 — Zero-shot vs Few-shot

pythonCopy
1from transformers import pipeline
2
3generator = pipeline(
4    "text-generation",
5    model="gpt2"
6)
7
8zero_prompt = """
9Classify:
10
11The movie was fantastic.
12"""
13
14few_prompt = """
15Positive → Great
16
17Negative → Bad
18
19Positive → Amazing
20
21Classify:
22
23The movie was fantastic.
24"""
25
26print(generator(
27    zero_prompt,
28    max_new_tokens=40
29)[0]["generated_text"])
30
31print(generator(
32    few_prompt,
33    max_new_tokens=40
34)[0]["generated_text"])

What You'll Learn

• Compare zero-shot and few-shot prompting.
• Observe how examples influence the model's responses.
• Understand when demonstrations improve performance.

Practice 2 — Structured Prompt Template

pythonCopy
1template = """
2Role:
3{role}
4
5Task:
6{task}
7
8Context:
9{context}
10
11Constraints:
12{constraints}
13
14Output Format:
15{format}
16"""
17
18prompt = template.format(
19    role="AI Tutor",
20    task="Explain Transformers",
21    context="Audience: Beginners",
22    constraints="Use simple language",
23    format="Markdown"
24)
25
26print(prompt)

What You'll Learn

• Build reusable prompt templates.
• Separate context, task, and constraints.
• Create prompts suitable for production systems.

Mini Project — Build an AI Assistant

pythonCopy
1from transformers import pipeline
2
3assistant = pipeline(
4    task="text-generation",
5    model="gpt2"
6)
7
8system_prompt = """
9You are an AI programming assistant.
10
11Rules:
12- Answer clearly.
13- Provide Python code when appropriate.
14- Explain concepts step by step.
15- Keep responses concise.
16"""
17
18user_question = """
19How do I reverse a list in Python?
20"""
21
22prompt = system_prompt + "\n\nUser:\n" + user_question + "\n\nAssistant:"
23
24response = assistant(
25    prompt,
26    max_new_tokens=120,
27    temperature=0.7
28)
29
30print(response[0]["generated_text"])

What You'll Learn

• Combine a system prompt with user input.
• Define consistent assistant behavior.
• Generate structured responses using a reusable prompt template.

Best Practices for Prompt Engineering

Module Summary

After completing this module, you will be able to:

• Explain the purpose and importance of prompt engineering.
• Distinguish between zero-shot, one-shot, and few-shot prompting.
• Understand the concepts behind Chain of Thought, Self Consistency, Tree of Thoughts, and ReAct.
• Design structured prompts with clear roles, tasks, context, constraints, and output formats.
• Create reusable prompt templates for different applications.
• Write effective system prompts that define assistant behavior.
• Build a simple AI assistant using prompt templates and Hugging Face Transformers.
• Choose the most appropriate prompting strategy based on the task and desired level of reasoning or tool use.

Next Module: Module 20 – Fine-Tuning & PEFT, where you'll learn transfer learning, Hugging Face Datasets, LoRA, QLoRA, PEFT, supervised fine-tuning (SFT), quantization, model evaluation, and deployment of custom LLMs.