Before vs. After Example

Before (Unclear): "Tell me about AI agents."

After (Clear): "Explain what AI agents are, how they work, and provide three examples of their practical applications in business settings. Include a brief explanation of how they differ from simple chatbots."

Before vs. After Example

Before (No Context): "Write code to process data."

After (With Context): "I'm a marketing analyst with basic Python knowledge. Write code to process CSV data containing customer purchase history. The data includes customer ID, purchase date, product ID, and amount. I need to calculate the average purchase value per customer and identify the top 10 customers by total spend."

Before vs. After Example

Before (Vague): "Write a blog post about AI safety."

After (Specific): "Write a 1000-word blog post about AI safety for a technical audience with some AI background. Structure it with an introduction, 3-4 main sections covering key risks (specifically including alignment problems and misuse), current mitigation approaches, and future research directions. Use a professional but accessible tone, include 2-3 real-world examples, and end with actionable takeaways for AI developers."

Prompt Iteration Example

Version 1: "Summarise this research paper."

Issue: Summary is too general and misses key findings.

Version 2: "Summarise this research paper, focusing on methodology and key findings."

Issue: Better, but lacks structure and is too verbose.

Version 3: "Create a structured summary of this research paper with these sections: 1) Research Question, 2) Methodology, 3) Key Findings, 4) Limitations. Keep each section concise with bullet points where appropriate."

Result: Well-structured, focused summary highlighting the most important aspects of the paper.

Example Implementation

import re

def extract_tool_calls(text):
    """Extract tool calls from text using regex."""
    # Improved regex to handle nested tags (non-greedy match)
    pattern = r'<(\w+)>(.*?)'
    tool_calls = []
    
    # Find all top-level tool calls
    matches = re.finditer(pattern, text, re.DOTALL)
    for match in matches:
        tool_name = match.group(1)
        tool_content = match.group(2).strip()
        
        # Extract parameters within this tool call
        params = {}
        param_pattern = r'<(\w+)>(.*?)'
        param_matches = re.finditer(param_pattern, tool_content, re.DOTALL)
        
        for param_match in param_matches:
            param_name = param_match.group(1)
            param_value = param_match.group(2).strip()
            # Basic handling for potential nested structures (more robust parsing needed for deep nesting)
            if '<' in param_value and '>' in param_value:
                # Could potentially parse nested structure here if needed
                pass # For simplicity, keep as string for now
            params[param_name] = param_value
        
        tool_calls.append({
            "tool": tool_name,
            "params": params
        })
    
    return tool_calls

# Example usage
text = """
I'll search for that information.


  latest AI research papers 2025
  5


Let me analyse these results for you.
"""

tool_calls = extract_tool_calls(text)
print(tool_calls)

Example: Dynamic Summary Prompt Generator

Input: User request ("Summarise this financial report for an executive")

Meta-Prompt: "Generate a detailed prompt for GPT-4 to summarise the provided text. The user is an executive. The summary should be concise, focus on key financial metrics, and highlight potential risks or opportunities."

Output Prompt (Generated): "You are an expert financial analyst. Summarise the following financial report specifically for a busy executive. Focus on: 1) Overall profit/loss trends, 2) Key revenue drivers, 3) Major expense categories, 4) Significant risks or opportunities identified. Keep the summary under 250 words using clear, non-technical language."

Persona Examples:

• "You are an expert Python programmer specializing in data analysis..."
• "You are a helpful customer support agent for a SaaS company..."
• "You are a creative copywriter crafting compelling marketing slogans..."
• "You are a meticulous editor focused on grammar, clarity, and style..."

LLM-as-Judge Example Prompt

You are an impartial evaluator. Evaluate the following AI-generated response based on the user query and specific criteria.

User Query:
[Insert User Query]

AI Response:
[Insert AI Response]

Evaluation Criteria:
1. Accuracy: Is the information factually correct?
2. Relevance: Does the response directly address the user query?
3. Clarity: Is the response easy to understand?
4. Completeness: Does the response cover all aspects of the query?

Instructions:
For each criterion, provide a score from 1 (Poor) to 5 (Excellent) and a brief justification.
Finally, provide an overall quality score from 1 to 5.

Evaluation:
Accuracy Score: [1-5]
Justification:
Relevance Score: [1-5]
Justification:
Clarity Score: [1-5]
Justification:
Completeness Score: [1-5]
Justification:

Overall Quality Score: [1-5]
Overall Justification: