Telusko Docs

Prompt Engineering is the practice of crafting effective prompts to obtain accurate, predictable, and specific outputs from AI language models. The quality of the response directly depends on the effectiveness of the prompt.

AI is like a very knowledgeable person who provides accurate answers when asked specific questions, but may give excessive or irrelevant information if not guided properly.

Why Prompt Engineering Matters

Bad Prompting	Good Prompting
Uncertain, unpredictable outputs	Predictable, specific outputs
Excessive or irrelevant information	Focused, relevant responses
Generic answers	Tailored, actionable results

Even as large language models improve, well-crafted prompts consistently produce better results than vague ones.

Four Components of a Good Prompt

Every effective prompt should include these key components:

Components_of_Prompt

Component	Purpose	Example
Instruction	What you want the AI to do	"Explain," "Generate," "Compare," "Summarize"
Input	The specific data or topic to work with	A code snippet, a paragraph, a question
Context	Background information to guide the response	User's expertise level, project details, constraints
Output Format	Expected structure of the response	JSON, table, bullet points, paragraph

Example

Instruction: Compare the following frontend frameworks
Input: React, Angular, Vue
Context: I am a junior developer choosing my first framework
Output: Present as a markdown table with columns for learning curve,
        community size, and job market demand

Adding more context results in better output — a well-contextualized prompt provides the AI with necessary information to generate accurate and relevant responses.

Prompt Engineering Techniques

Zero-Shot Prompting

The simplest approach — a clear, concise instruction without any examples.

Aspect	Detail
How it works	Relies solely on clear instructions to guide the response
When to use	Simple questions, quick answers, straightforward tasks
Limitation	Least controlled; output format may vary

Example:

What is a smart contract in blockchain?

Few-Shot Prompting

Providing examples of the desired output format along with the prompt, so the model learns the expected pattern.

Aspect	Detail
How it works	Give 2-3 examples of the desired output before asking the actual question
When to use	When you need a specific format, structure, or style
Benefit	Guarantees consistent, structured outputs matching the desired format

Example — Generating Commit Messages:

Generate a commit message for the following change.

Examples:
- Change: Added user authentication → Commit: feat: add user authentication with JWT tokens
- Change: Fixed navbar alignment → Commit: fix: resolve navbar alignment issue on mobile
- Change: Updated README → Commit: docs: update README with installation instructions

Now generate for:
- Change: Added login button to the homepage

Output: feat: add login button to homepage

Few-shot prompting establishes a pattern that the AI follows, ensuring consistent and predictable outputs.

Chain of Thought (CoT) Prompting

Forces the AI to break down complex problems into step-by-step reasoning, similar to showing work in mathematics.

Aspect	Detail
How it works	Ask the model to "think step by step" or show its reasoning
When to use	Complex calculations, logical problems, multi-step reasoning
Benefit	Increases accuracy and allows verification of the logic path

Example:

A store has 45 apples. They sell 12 in the morning and receive
a shipment of 30 in the afternoon. How many apples do they have now?

Think step by step.

Output:

Step 1: Start with 45 apples
Step 2: Sell 12 → 45 - 12 = 33 apples
Step 3: Receive 30 → 33 + 30 = 63 apples
Answer: 63 apples

Adding "think step by step" slows down the model's response, improving accuracy by forcing it to reason through each stage rather than jumping to an answer.

Role Prompting

Assigning a specific persona or role to the AI, causing it to adjust technical depth, terminology, and examples to match that role's expertise.

Aspect	Detail
How it works	Prefix the prompt with a role assignment
When to use	When you need responses tailored to a specific expertise level or perspective
Benefit	Transforms the depth and style of responses

Example:

You are a senior blockchain architect. Explain smart contracts
to a junior developer using relatable analogies.

The AI will adjust its response to be technically informed but accessible, using analogies appropriate for a beginner.

Structured Output Prompting

Specifying the exact output format (JSON, CSV, table, list) to get parseable, predictable data.

Aspect	Detail
How it works	Explicitly state the desired format in the prompt
When to use	AI-powered applications that need machine-readable data
Benefit	Predictable, parseable output rather than free-form text

Example:

List the top 3 JavaScript frameworks with their release year
and primary use case. Return the result as JSON.

Output:

[
  { "name": "React", "year": 2013, "useCase": "UI component libraries" },
  { "name": "Angular", "year": 2016, "useCase": "Enterprise SPAs" },
  { "name": "Vue", "year": 2014, "useCase": "Progressive web apps" }
]

Structured output prompting is critical for AI-powered applications that need to parse and process AI responses programmatically.

Prompt_Technique_Comparison

Context Optimization

Embedding contextual information about the user's background, goals, and expertise level enables AI to generate tailored responses rather than generic answers.

Context Element	Impact
User's expertise level	Adjusts complexity of explanation
Project constraints	Focuses on relevant solutions
Desired outcome	Aligns response with goals
Industry/domain	Uses appropriate terminology

Summary

Prompt Engineering is a critical skill for AI Engineers, the quality of output directly depends on the quality of the prompt.
A good prompt has four components: Instruction, Input, Context, and Output Format.
Zero-shot is simplest but least controlled; few-shot ensures format consistency through examples.
Chain of Thought improves accuracy on complex problems by forcing step-by-step reasoning.
Role prompting tailors responses to specific expertise levels and perspectives.
Structured output is essential for building AI-powered applications that need predictable, parseable data.
Always provide clear instructions, relevant context, and specify the expected output format for best results.

Written By: Muskan Garg

Prompt Engineering

On this page