Prompt

Engineering
Introduction
Prompt Engineering
Definition

“Prompt engineering is the process of crafting

and refining input instructions (prompts)
given to large language models (LLMs) to
achieve the desired output. It involves
understanding the model's capabilities and
limitations to optimize its performance for
specific tasks.”
Prompt Engineering
Objective

The primary goal is to design

prompts that elicit accurate,
relevant, and context-aware
responses from an LLM. This
includes maximizing output
quality while minimizing
ambiguity or irrelevant
information.
Prompt Engineering
Components of effective prompts

1. Clear and specific instructions.

2. Sufficient context for complex queries.
3. Use of examples or templates for guidance.
4. Explicit tone or format requirements (e.g., "Use bullet points for clarity.").
5. Clear Output Constraints
6. Defined Target Audience (if applicable)
7. Conditional Instructions
8. Clarification Requests
9. Time Sensitivity
10. Bias and Sensitivity Considerations
TYPES OF PROMPT
ENGINEERING
01. Zero-Shot Learning

02. Few-Shot Learning

03. Chain-of-Thought (CoT)

04. Tree-of-Thought (ToT)

05. Retrieval-Augmented
Generation(RAG)
 Zero-Shot
01 Learning
01. Zero-Shot Learning
Prompt engineering

1. Zero-Shot Learning refers to the model’s ability to make predictions on tasks it has
not seen during training.
2. The model utilizes learned representations from similar tasks or categories to make
inferences without requiring labeled examples.
3. It leverages semantic information such as textual descriptions or attributes to
recognize and classify unseen classes.
4. Zero-shot learning is especially useful when it's difficult or costly to collect labeled
data for all possible categories.
5. This type of learning relies on generalization capabilities that allow the model to
extrapolate from its training data to unfamiliar situations.
6. Commonly used in tasks like classification, text generation, and even image
recognition, without explicit task-specific training.
7. It’s critical in situations where the model needs to handle evolving or dynamic
categories, such as real-time applications in industries like e-commerce.
01. Zero-Shot Learning
Prompt engineering

Here’s an example:

"I need to expand the following brief idea into a comprehensive and
engaging blog post of approximately 1,000 words. Break down the topic
into multiple sections, each with a clear heading. Include an introduction to
set the context, body paragraphs that provide depth and actionable
insights, and a conclusion that reinforces the key takeaways. Use an
engaging, conversational tone, and incorporate relevant examples to make
the content more relatable. Here’s the idea to expand:

'The importance of time management in remote work.'"

Zero-Shot Learning
Prompt engineering
02 Few-Shot Learning
02. Few-Shot Learning
Prompt engineering
02. Few-Shot Learning
Prompt engineering

1. Few-Shot Learning is a learning paradigm where the model is provided with a very
small number of labeled examples for a new task.
2. Unlike traditional machine learning that requires large datasets, FSL aims to
perform well with just a few data points.
3. It generally works by leveraging pre-trained models on large datasets and
fine-tuning them with minimal task-specific data.
4. FSL is built on the concept of transfer learning, where knowledge from one domain
is transferred to solve problems in another.
5. The model uses meta-learning approaches to learn how to adapt quickly to new
tasks with few examples.
6. This paradigm is useful for problems where labeling data is expensive or when
there are limitations in gathering large amounts of data.
7. Applications include language modeling, image classification, and medical
diagnosis, where datasets for certain conditions or topics may be scarce.
Few-Shot Learning
Prompt engineering

Here’s an example:

"I need to expand the following brief idea into a comprehensive and
engaging blog post of approximately 1,000 words. Break down the topic
into multiple sections, each with a clear heading. Include an introduction to
set the context, body paragraphs that provide depth and actionable
insights, and a conclusion that reinforces the key takeaways. Use an
engaging, conversational tone, and incorporate relevant examples to make
the content more relatable. Here’s the idea to expand:

'The importance of time management in remote work.'"

03 Chain-of-Thought (CoT)
03. Chain-of-Thought (CoT)
Prompt engineering

1. Chain-of-Thought (CoT) is a reasoning technique that involves breaking down

complex problems into a series of smaller, manageable steps.
2. It encourages the model to "think" step-by-step, improving its ability to reason
through problems and arrive at more accurate solutions.
3. CoT is often used in tasks that require logical reasoning, such as mathematical
problem solving and decision-making processes.
4. The process can involve both symbolic reasoning and logical deduction, where
each thought leads directly to the next.
5. It’s commonly applied in natural language processing tasks, particularly those
involving arithmetic, complex question answering, or multi-step reasoning.
6. CoT enables better interpretability of the model’s decision-making process by
showing the intermediate reasoning steps.
7. By making reasoning more explicit, it improves accuracy, reduces errors, and
enhances the model's transparency.
03. Chain-of-Thought (CoT)
Prompt engineering

Here’s an example:

"I need to expand the following brief idea into a comprehensive and
engaging blog post of approximately 1,000 words. Break down the topic
into multiple sections, each with a clear heading. Include an introduction to
set the context, body paragraphs that provide depth and actionable
insights, and a conclusion that reinforces the key takeaways. Use an
engaging, conversational tone, and incorporate relevant examples to make
the content more relatable. Here’s the idea to expand:

'The importance of time management in remote work.'"

03. Chain-of-Thought (CoT)
Prompt engineering
04 Tree-of-Thought (ToT)
04. Tree-of-Thought (ToT)
Prompt engineering

1. Tree-of-Thought (ToT) builds upon Chain-of-Thought by using a hierarchical

structure of reasoning, branching out like a tree.
2. It allows the model to evaluate multiple pathways of reasoning before converging
on the final answer.
3. The decision process in ToT is non-linear, where each thought can branch out in
multiple directions, forming a tree-like structure.
4. This enables deeper reasoning by considering alternative outcomes or decisions
and refining the solution by evaluating several options.
5. ToT is particularly useful in problems requiring exploration of multiple possibilities,
such as strategic planning or hypothesis testing.
6. It can be used in multi-step tasks that involve nested decision points or complex
logical conclusions.
7. ToT provides an enhanced form of structured reasoning, where each node of the
tree represents a different reasoning step or decision point.
04. Tree-of-Thought (ToT)
Prompt engineering

Here’s an example:

"I need to expand the following brief idea into a comprehensive and
engaging blog post of approximately 1,000 words. Break down the topic
into multiple sections, each with a clear heading. Include an introduction to
set the context, body paragraphs that provide depth and actionable
insights, and a conclusion that reinforces the key takeaways. Use an
engaging, conversational tone, and incorporate relevant examples to make
the content more relatable. Here’s the idea to expand:

'The importance of time management in remote work.'"

04. Tree-of-Thought (ToT)
Prompt engineering
 Retrieval-Augmented
05 Generation (RAG)
05. Retrieval-Augmented Generation (RAG)
Prompt engineering

1. Retrieval-Augmented Generation (RAG) is a hybrid approach that combines

traditional retrieval-based methods with generative models.
2. The model first retrieves relevant information from an external knowledge base or
a pre-indexed database.
3. Once the relevant information is retrieved, it is used to inform and guide the
generative model to create an output, such as text or an answer.
4. RAG combines the advantages of both retrieval systems (fast and accurate for
finding relevant facts) and generative systems (creating novel responses).
5. This approach allows the model to produce more accurate and contextually
relevant outputs by leveraging external knowledge.
6. RAG is particularly effective in domains where up-to-date or specialized
information is needed, like customer service, question answering, or medical
applications.
7. It can help address the limitations of generative models by grounding the
generated content in factual, retrieved data, thus improving reliability and
accuracy.
05. Retrieval-Augmented Generation (RAG)
Prompt engineering

Here’s an example:

"I need to expand the following brief idea into a comprehensive and
engaging blog post of approximately 1,000 words. Break down the topic
into multiple sections, each with a clear heading. Include an introduction to
set the context, body paragraphs that provide depth and actionable
insights, and a conclusion that reinforces the key takeaways. Use an
engaging, conversational tone, and incorporate relevant examples to make
the content more relatable. Here’s the idea to expand:

'The importance of time management in remote work.'"

05. Retrieval-Augmented Generation (RAG)
Prompt engineering
 Prompt
Engineering
Real world applications
 Summarization
01 Condenses information while retaining key
details and meaning.
01. Summarization
Prompt engineering

1. Summarization involves condensing a large piece of content into a shorter version

while retaining its key points and meaning.
2. It can be applied to text, speech, or even video, creating a concise representation of
the original content.
3. Techniques include extractive summarization (selecting key sentences) and
abstractive summarization (paraphrasing).
4. Summarization is commonly used in news aggregation, academic research, and
content curation.
5. It aids in quickly digesting large volumes of information, saving time and
enhancing productivity.
6. The challenge lies in maintaining the accuracy and richness of the original content
while reducing its length.
01. Summarization
Prompt engineering

Here’s an example:

“I have a long document that needs to be summarized for a professional

audience. Please focus on the key points and actionable insights while
removing unnecessary details. Your response should not exceed 200 words.
Use a formal tone, and make sure to preserve the original intent and
meaning of the document. Here’s the text to summarize:

[Paste text here].”

01. Summarization
Prompt engineering
 Expansion
02 Converts brief input into detailed, creative, or
technical outputs.
02. Expansion
Prompt engineering

1. Expansion refers to elaborating on a given idea or concept to provide more

detailed information or context.
2. It is often used in creative writing, marketing copy, and educational materials to
make the content more comprehensive.
3. The expansion process can involve adding examples, explanations, or breaking
down complex ideas.
4. This is useful in generating content for blog posts, articles, or product descriptions.
5. In natural language processing, it often includes rephrasing and generating more
descriptive or informative text.
6. Expansion helps to increase the depth and richness of content, improving user
engagement and clarity.
02. Expansion
Prompt engineering

Here’s an example:

"I need to expand the following brief idea into a comprehensive and
engaging blog post of approximately 1,000 words. Break down the topic
into multiple sections, each with a clear heading. Include an introduction to
set the context, body paragraphs that provide depth and actionable
insights, and a conclusion that reinforces the key takeaways. Use an
engaging, conversational tone, and incorporate relevant examples to make
the content more relatable. Here’s the idea to expand:

'The importance of time management in remote work.'"

02. Expansion
Prompt engineering
 Translation
03 Converts text between languages while
maintaining context, tone, and cultural relevance.
03. Translation
Prompt engineering

1. Translation involves converting text from one language to another while

maintaining the meaning and nuances of the original content.
2. Modern AI-powered translation systems use deep learning models to improve
accuracy and fluency in translated text.
3. This application is vital in breaking down language barriers in communication,
marketing, and international business.
4. Neural machine translation (NMT) models are commonly used for producing more
natural translations compared to earlier rule-based methods.
5. It has become integral in tools like Google Translate, making it easier for users to
understand foreign languages.
6. Challenges include maintaining cultural nuances and context, especially with
idiomatic expressions and domain-specific terms.
03. Translation
Prompt engineering

Here’s an example:

"Translate the following text from English to Spanish. Ensure that the
translation is culturally relevant and contextually accurate. Retain the tone
of the original text, whether it is formal, conversational, or technical. If there
are idioms or phrases that don’t translate directly, adapt them to match the
target language’s equivalent expressions. Here’s the text to translate:

[Insert text here]."

03. Translation
Prompt engineering
 Classification
04 Categorizes or labels text based on specific
criteria.
04. Classification
Prompt engineering

1. Classification involves assigning input data to predefined categories or classes

based on its features.
2. This application is commonly used in tasks like sentiment analysis, spam detection,
and image classification.
3. Machine learning models like decision trees, support vector machines, and neural
networks are often used for classification tasks.
4. It helps in organizing and interpreting large datasets in a way that is easy to
analyze and act upon.
5. In text, classification models can categorize documents, emails, or social media
posts based on content.
6. The accuracy of classification depends on the quality of training data and the
complexity of the categories.
04. Classification
Prompt engineering

Here’s an example:

"Classify the following reviews into one of these categories: Positive,

Negative, or Neutral. For each review, provide a brief explanation for why
you placed it in that category. Pay attention to the tone, wording, and
overall sentiment of the review. If any reviews seem ambiguous, explain why
and suggest an alternate way to interpret them. Here are the reviews:

[Insert review here]

[Insert review here]"

04. Classification
Prompt engineering
 Generate
Knowledge
05 Synthesizes and presents new information or
insights.
05. Generate Knowledge
Prompt engineering

1. Generating knowledge involves creating new insights or content based on existing

information or patterns.
2. AI models are used to identify gaps in current knowledge and suggest new
hypotheses or ideas.
3. This is used in fields like scientific research, healthcare, and business to uncover
new discoveries or patterns.
4. It leverages large datasets, existing literature, and expert knowledge to propose
novel solutions or answers.
5. Knowledge generation tools can automate literature reviews, summarize research,
and even identify emerging trends.
6. The challenge is ensuring the generated knowledge is accurate, relevant, and
aligned with human understanding.
05. Generate Knowledge
Prompt engineering

Here’s an example:

"Explain the following concept in simple terms for someone with no prior
knowledge of the topic. Structure your response into three sections: (1) An
introduction that defines the concept, (2) A main body with detailed
explanations and real-world examples, and (3) A conclusion summarizing its
importance. The response should be clear, concise, and no longer than 500
words. Here’s the concept: 'Quantum computing and its applications in
healthcare and finance.'"
05. Generate Knowledge
Prompt engineering
 Synthetic Dataset
06 Creation
Generates realistic, diverse, and customizable
data for training AI or testing systems.
06. Synthetic Dataset Creation
Prompt engineering

1. Synthetic dataset creation involves generating artificial data that mimics

real-world data for training machine learning models.
2. It’s used when real data is scarce, expensive, or involves privacy concerns (such as
medical data).
3. AI-generated synthetic datasets can be used to train models in areas like image
recognition, natural language processing, and autonomous driving.
4. The quality of synthetic data must match real data to avoid introducing biases or
inaccuracies in the model.
5. Generating diverse and representative synthetic data helps to improve the model’s
generalization capabilities.
6. Challenges include maintaining realism, diversity, and avoiding overfitting due to
data similarities.
06. Synthetic Dataset Creation
Prompt engineering

Here’s an example:

"Generate a synthetic dataset of 20 rows to simulate email spam

classification. The dataset should include the following columns: Sender
Name, Subject Line, Email Content, and Spam (Yes/No). Provide a realistic
mix of spam and non-spam examples, ensuring diversity in subject lines
and content. For spam emails, include phrases commonly flagged as spam
(e.g., 'Congratulations, you won!'). For non-spam, use neutral or professional
content. Present the dataset in a clean table format."
06. Synthetic Dataset Creation
Prompt engineering
 Retrieval-Augmen
07 ted Generation
Combines live or static external data retrieval
with text generation for informed responses.
07. Retrieval-Augmented Generation
(RAG)
Prompt engineering

1. RAG combines real-time data retrieval with generative models to enhance

response accuracy.
2. The system retrieves relevant information from external sources, then generates
context-aware output based on this data.
3. It allows for more informed responses, making it ideal for dynamic, real-time
applications.
4. In e-commerce, RAG can provide up-to-date product recommendations or pricing
information.
5. It improves customer support by generating accurate answers based on live FAQ
databases.
6. RAG is critical in research, where it can synthesize insights from academic papers
or other knowledge sources.
07. Retrieval-Augmented Generation
(RAG)
Prompt
Here’s anengineering
example:

"Using the following context, generate a detailed response to the question

provided below. Structure your answer with an introduction, key points
supported by data from the context, and a conclusion. If any information is
unavailable in the context, explicitly state that. Ensure your response is
accurate and avoids speculation.

Context: [Insert source or text here].

Question: 'What are the economic implications of adopting renewable

energy globally?'"
07. Retrieval-Augmented Generation
(RAG)
Prompt engineering
08 Code Generation &
Debugging
Helps developers write, debug, and
document code.
08. Code Generation and Debugging
Prompt engineering

1. Code generation automates the process of writing code, especially for repetitive or
boilerplate tasks.
2. It assists in bug fixing by detecting errors in code and suggesting possible
solutions or corrections.
3. Refactoring helps optimize and simplify existing code, improving performance and
readability.
4. Documentation generation enables developers to automatically create comments
or guides for code.
5. It speeds up software development by reducing manual coding effort and
enhancing productivity.
6. This application is useful across programming languages, reducing the burden on
developers and improving code quality.
08. Code Generation and Debugging
Prompt engineering

Here’s an example:

"Write a Python script to scrape product data from an e-commerce website

like Amazon using the BeautifulSoup library. The script should fetch details
like product name, price, and rating. Include robust error handling for
scenarios like network failures or missing data. Add comments to explain
each part of the code. After writing the script, provide a brief explanation of
how it works and how it can be adapted to other websites."
08. Code Generation and Debugging
Prompt engineering
09 Question
Answering (QA)
Directly answers user queries based on
context or external data.
09. Question Answering (QA)
Prompt engineering

1. QA systems answer user queries by retrieving information from structured or

unstructured data sources.
2. The system uses context to provide accurate, relevant answers to both direct and
complex questions.
3. It is applied in customer support, where it automates responses to common
product inquiries.
4. QA can be used in education to help students by answering homework questions
or explaining concepts.
5. Healthcare applications leverage QA to explain symptoms, medical terminology, or
treatment options to patients.
6. E-commerce platforms use QA to assist customers with product details, pricing, or
availability.
09. Question Answering (QA)
Prompt engineering

Here’s an example:

"Answer the following question in a detailed yet concise manner. Structure

your response into three parts: (1) A direct answer to the question, (2)
Supporting evidence or examples to back up your answer, and (3) A
summary or actionable insights if applicable. Avoid vague responses and
cite examples when possible.

Question: 'How does inflation impact small businesses?'"

09. Question Answering (QA)
Prompt engineering
10 Conversational
Agents
Enables human-like interactions for support
or entertainment.
10. Conversational Agents
Prompt engineering

1. Conversational agents, or chatbots, simulate human-like conversations for

customer support, sales, or entertainment.
2. They engage users in real-time, understanding and responding to user queries
with contextually relevant replies.
3. In customer service, they handle routine queries, freeing human agents for more
complex issues.
4. They provide entertainment through interactive storytelling, role-playing games, or
virtual characters.
5. Conversational agents are used in mental health applications, offering support and
guidance to users.
6. They enhance sales by guiding users through product selection, offering
recommendations, and completing transactions.
10. Conversational Agents
Prompt engineering

Here’s an example:

"Simulate a customer service chatbot for an e-commerce platform.

Respond to the following query as a polite and professional representative.
Be concise but thorough in explaining the steps involved. Provide options
for follow-up queries or additional help.

Query: 'I received a damaged product. What are the steps to initiate a
return?'"
10. Conversational Agents
Prompt engineering
11 Content
Moderation
Filters or flags inappropriate, offensive, or
policy-violating content.
11. Content Moderation
Prompt engineering

1. Content moderation uses AI to filter and flag inappropriate or harmful content in

text, images, or videos.
2. It ensures that online platforms adhere to community guidelines and prevent the
spread of harmful material.
3. Social media platforms use content moderation to detect hate speech, spam, and
misinformation.
4. Forums and e-commerce websites employ moderation to maintain safe, respectful
environments.
5. Automated moderation reduces manual workload, but still requires human
oversight for edge cases.
6. It helps protect the reputation of businesses, social platforms, and brands by
enforcing policies.
11. Content Moderation
Prompt engineering

Here’s an example:

"Analyze the following social media comments and identify any that violate
community guidelines for hate speech, spam, or inappropriate language.
For each flagged comment, provide a brief explanation of the violation and
suggest an appropriate moderation action (e.g., warning, removal, or ban). If
a comment is borderline, explain your reasoning.

Comments:

1. [Insert comment here]

2. [Insert comment here]"
11. Content Moderation
Prompt engineering
 Personalization
12 Customizes content, recommendations, or
experiences for individual users.
12. Personalization
Prompt engineering

1. Personalization customizes user experiences based on preferences, behaviors, or

demographics.
2. It improves user engagement by tailoring content, recommendations, and
interactions.
3. In e-commerce, personalized product recommendations increase conversions and
customer satisfaction.
4. Education platforms use personalization to adapt lesson plans or courses based on
student performance.
5. Streaming services recommend TV shows or movies based on user viewing history
and tastes.
6. Personalized healthcare plans offer tailored fitness or diet recommendations for
individuals.
12. Personalization
Prompt engineering

Here’s an example:

"Create personalized product recommendations for a user who recently

searched for fitness equipment and healthy recipes. Provide 5 product
suggestions that are relevant to their interests. For each product, include a
brief explanation of why it was recommended and how it fits their needs.
Present your response in a professional and user-friendly tone."
12. Personalization
Prompt engineering
13 Knowledge Base
Curation
Automatically structures and updates
knowledge bases.
13. Knowledge Base Curation
Prompt engineering

1. Knowledge base curation involves organizing and updating knowledge resources

to ensure accuracy and relevance.
2. It streamlines the process of categorizing and storing information for easy access
and retrieval.
3. Corporate knowledge bases house internal documents, guides, and best practices
for teams.
4. Customer support teams build FAQs and troubleshoot documentation from
historical interactions.
5. In research, knowledge bases summarize and categorize academic findings for
easier reference.
6. Legal organizations maintain case law databases, providing quick access to
relevant precedents and regulations.
13. Knowledge Base Curation
Prompt engineering

Here’s an example:

"Using the raw customer support data provided below, create a

well-structured FAQ document. Organize the content into clear sections
with appropriate headings. Each question should have a concise answer
written in simple language. If any information is unclear, make logical
assumptions and highlight them as placeholders. Ensure the final FAQ is
professional and easy to navigate.

Data: [Insert raw data or conversation logs here]."

13. Knowledge Base Curation
Prompt engineering
THANKS!
Do you have any questions?

[email protected]
+91 620 421 838
ALTERNATIVE
RESOURCES
Written:
● Prompt Engineering Guide
● Change the output of your prompt |
Microsoft Learn
● Prompts Masterclass: Output Formatting —
JSON #5 - AI Mind
● Prompt Engineering ~ LLMs at Work - Vlad
Rișcuția
● Prompt Engineering: JSON or YAML? Why
YAML has the Edge

Videos:
● An Introduction to Artificial Intelligence (AI)
for Beginners | SparX by Mukesh Bansal
● Prompt Engineering 2024 Full course |
Prompt engineering course | ChatGPT
Prompts