IndianTechnoEra: Real-World Usage Examples of LLMs | Prompt Engineering: Master the Language of AI

Chapter 6: Prompt Engineering Course

This module explores practical applications of Large Language Models (LLMs) across various domains. You'll learn how to craft effective prompts for common real-world tasks, with examples and interactive demonstrations.

1. Structured Data Extraction

LLMs can transform unstructured text into structured formats like JSON, CSV, or tables. This is particularly useful for data processing pipelines.

Example Prompt:

Extract the product details from the following text as JSON with fields for name, price, and features:

"The Quantum X900 laptop features a 15.6" 4K display, 32GB RAM, and 1TB SSD storage. It's priced at $1,299.99 and comes with a 2-year warranty."

Output:

{
  "name": "Quantum X900 laptop",
  "price": 1299.99,
  "features": [
    "15.6\" 4K display",
    "32GB RAM",
    "1TB SSD storage",
    "2-year warranty"
  ]
}

Python Implementation:

from transformers import pipeline

# Initialize the text-to-json pipeline
extractor = pipeline("text2text-generation", model="google/flan-t5-large")

# Define the prompt and input text
prompt = "Extract the product details as JSON with name, price, and features:"
text = "The Quantum X900 laptop features a 15.6\" 4K display..."

# Generate the structured output
result = extractor(prompt + text)
print(result['generated_text'])

2. Inferring: Sentiment Analysis & Summarization

LLMs can analyze text for sentiment, extract key topics, and create concise summaries—valuable for processing customer feedback or research papers.

Sentiment Analysis Prompt:

Analyze the sentiment of this product review as positive, neutral, or negative:

"The battery life is excellent, lasting a full day of heavy use. However, the keyboard feels cheap and the touchpad is unresponsive at times."

Output:

Sentiment: Mixed (positive and negative aspects)

Positive aspects: Battery life
Negative aspects: Keyboard quality, touchpad responsiveness

Summarization Prompt:

Summarize the following article in 3 bullet points:

[Article text about recent advancements in renewable energy...]

Output:

• New solar panel design achieves 25% higher efficiency
• Breakthrough in battery storage reduces costs by 40%
• Global renewable energy capacity grew by 15% in 2024

3. Writing Professional Emails

LLMs can help craft well-structured emails with specific tones, saving time on business communication.

Prompt:

Write a formal email to a client named Mr. Johnson, explaining a 2-week delay in the project timeline due to unforeseen technical challenges. Assure them we're allocating additional resources and will deliver by the new deadline of June 15th.

Output:

Subject: Update on Project Timeline

Dear Mr. Johnson,

I hope this message finds you well. I'm writing to inform you that we've encountered unforeseen technical challenges that require us to adjust the project timeline. After careful assessment, we've determined that we'll need approximately two additional weeks to resolve these issues while maintaining our quality standards.

Please rest assured that we've already allocated additional resources to address these challenges, and our team is working diligently to complete the project by the new deadline of June 15th.

We sincerely apologize for this adjustment and appreciate your understanding. Should you have any questions or require additional information, please don't hesitate to contact me directly.

Best regards,
[Your Name]
[Your Position]

Prompt:

Write a persuasive email to potential customers about our new productivity software, emphasizing time savings (average 10 hours/week), seamless integration, and the limited-time 30% discount offer ending May 31st.

Output:

Subject: Save 10 Hours Weekly + 30% Off Ends Soon

Hi [First Name],

Are you tired of wasting precious hours on repetitive tasks? Our new ProductivityPro software helps professionals like you save an average of 10 hours every week by automating routine work and seamlessly integrating with your existing tools.

For a limited time, we're offering 30% off all annual subscriptions—but this special discount ends May 31st. This is your chance to:

• Regain 40+ hours each month
• Work smarter with our AI-powered automation
• Connect all your tools in one dashboard

Join thousands of professionals who have transformed their workdays. Click below to claim your discount before it's gone:

[Get ProductivityPro Now]

Best,
[Your Name]

Prompt:

Write a polite follow-up email to a colleague who hasn't responded to your previous request for feedback on the project proposal. Keep it friendly but professional.

Output:

Subject: Gentle Reminder: Feedback Request

Hi [Colleague's Name],

I hope you're doing well! I wanted to follow up on my previous email from [date] regarding your feedback on the [Project Name] proposal. I know you're busy, but your insights would be incredibly valuable as we finalize the plan.

If you could share your thoughts by [specific date if applicable], it would help us stay on track. Of course, if you need more time or have any questions, please let me know.

Thanks so much for your help with this!

Best regards,
[Your Name]

4. Coding Assistance

LLMs excel at explaining, generating, and debugging code across multiple programming languages.

Code Generation Prompt:

Write a Python function that takes a list of numbers and returns a dictionary with the mean, median, and mode. Include docstring and type hints.

Output:

from typing import List, Dict
from statistics import mean, median, mode

def calculate_stats(numbers: List[float]) -> Dict[str, float]:
  """Calculate basic statistics from a list of numbers.

  Args:
    numbers: List of numeric values

  Returns:
    Dictionary with 'mean', 'median', and 'mode' keys
  """
  try:
    mode_val = mode(numbers)
  except StatisticsError:
    mode_val = None

  return {
    "mean": mean(numbers),
    "median": median(numbers),
    "mode": mode_val
  }

Debugging Prompt:

Explain the error in this Python code and suggest a fix:

def calculate_discount(price, discount):
  return price - (price * discount)

# When called with calculate_discount(100, 0.2) it works,
# but with calculate_discount(100, 20) it gives wrong result

Output:

# The issue is inconsistent discount parameter handling.
# The function expects discount as decimal (0.2 for 20%),
# but sometimes gets percentage (20 for 20%).

# Solution 1: Document expectation and validate input
def calculate_discount(price: float, discount: float) -> float:
  """Calculate price after discount.
  Args:
    discount: Decimal (e.g., 0.2 for 20% off)
  """
  if discount > 1: # Assume percentage was passed
    discount /= 100
  return price * (1 - discount)

5. Study Buddy & Learning Assistance

LLMs can simplify complex topics, generate study materials, and create practice questions.

Simplification Prompt:

Explain quantum computing to a high school student using simple analogies and avoid technical jargon. Focus on the key concepts of qubits, superposition, and entanglement.

Output:

Quantum computing is like having a magical notebook that can be in multiple states at once. While regular computers use bits (like light switches that are either ON or OFF), quantum computers use qubits that can be ON, OFF, or both at the same time—this is called superposition.

Imagine spinning a coin—while it's spinning, it's not just heads or tails, but in a mix of both states. Qubits work similarly until you "look" at them (measure them), when they settle into one state.

Entanglement is like having two magical coins that are linked—no matter how far apart they are, if you spin one and it lands heads, the other will instantly land tails. This spooky connection helps quantum computers solve certain problems much faster than regular computers.

Quiz Generation Prompt:

Create a 5-question multiple choice quiz about basic machine learning concepts for beginners. Include one question about supervised vs. unsupervised learning, one about neural networks, one about overfitting, one about feature engineering, and one about evaluation metrics. Provide the correct answers.

Output:

1. What is the main difference between supervised and unsupervised learning?

• A) Supervised learning uses labeled data, unsupervised doesn't
• B) Supervised learning is faster
• C) Unsupervised learning always gives better results

2. What are neural networks loosely inspired by?

• A) The structure of atoms
• B) The human brain's neurons
• C) Computer processors

3. Overfitting occurs when a model:

• A) Performs well on training data but poorly on new data
• B) Is too simple to capture patterns
• C) Has too few parameters

Correct answers: 1-A, 2-B, 3-A

6. Designing Chatbots

LLMs can power conversational agents with predefined behaviors and personalities.

Chatbot Prompt:

You are a friendly customer support bot for an e-commerce store called "TechGadgets". Your role is to:
- Greet customers warmly
- Answer questions about products, shipping, and returns
- Escalate complex issues to human agents
- Maintain a helpful but professional tone

The customer says: "Hi, I ordered a wireless charger 3 days ago but haven't received any shipping update. Can you help?"

Output:

"Hello! Thanks for reaching out to TechGadgets support. I'd be happy to check on your order. Could you please share your order number so I can look up the details? Our standard shipping usually takes 3-5 business days, but I can check the current status for you."

Python Implementation:

from transformers import pipeline

# Initialize the chatbot with our system prompt
chatbot = pipeline("conversational", model="facebook/blenderbot-400M-distill")

# System message defines the bot's behavior
system_prompt = """You are a friendly customer support bot for TechGadgets..."""

# Conversation history
conversation = [
  {"role": "system", "content": system_prompt},
  {"role": "user", "content": "Hi, I ordered a wireless charger..."}
]

# Generate response
response = chatbot(conversation)
print(response['generated_text'])

7. Content Generation

LLMs can assist with creating blog posts, social media content, and other marketing materials.

Blog Post Prompt:

Write a 300-word blog post introduction about the benefits of remote work for both employees and employers. Focus on productivity, work-life balance, and cost savings. Use a friendly, engaging tone and include statistics where relevant.

Output:

The Remote Work Revolution: Why Everyone Wins

The traditional 9-to-5 office model is undergoing a dramatic transformation. According to a 2024 Global Workplace Report, 72% of knowledge workers now operate remotely at least part-time—and both employees and employers are reaping the benefits.

For employees, remote work offers unparalleled flexibility. A Stanford study found that remote workers reported 25% less stress and 50% lower turnover rates. The elimination of commutes (saving the average worker 200+ hours annually) translates directly into better work-life balance and more time for family, hobbies, and self-care.

Employers benefit too. Companies report 15-20% productivity boosts from remote teams, along with significant real estate savings. Tech giant Dell saved $12 million annually by reducing office space, while maintaining employee satisfaction scores above 90%.

8. Data Analysis Support

LLMs can help interpret data, generate insights, and even suggest visualizations.

Analysis Prompt:

Analyze this sales data and identify 3 key insights and 2 recommendations:

- Q1 Revenue: $1.2M (15% growth YoY)
- Q2 Revenue: $1.1M (8% growth YoY)
- Customer acquisition cost increased from $50 to $75
- Retention rate dropped from 85% to 78%
- New product line contributed 25% of Q2 revenue

Output:

Key Insights:

1. Revenue growth is slowing despite new product success (25% contribution)
2. Increasing acquisition costs and declining retention suggest customer satisfaction issues
3. The new product line is performing well but may be cannibalizing existing products

Recommendations:

1. Investigate reasons for retention drop (survey customers, analyze support tickets)
2. Allocate more budget to retain existing customers rather than acquire new ones

9. Prompt Engineering Workflow

Effective prompt engineering follows an iterative process of refinement and testing.

Best Practices:

• Start specific: Clearly define the task and desired output format
• Provide examples: Show the model what good output looks like
• Use constraints: Limit response length or format when needed
• Iterate: Test multiple prompt variations to find what works best
• Evaluate: Check outputs for accuracy, completeness, and relevance

Prompt Refinement Example:

Initial Prompt:

Tell me about machine learning

Refined Prompt:

Explain supervised machine learning to a business executive in 3 paragraphs. Focus on practical applications in customer analytics and use analogies from everyday life. Avoid mathematical terminology.

Further Refined:

You are explaining supervised machine learning to a retail company's CEO who has no technical background. Structure your response as:
1) Core concept analogy (like teaching a child with examples)
2) Two business use cases from their industry (customer segmentation and demand forecasting)
3) Implementation requirements (data needed, timeline, expected outcomes)
Keep each section under 100 words.