#!/usr/bin/env python3
"""
Example usage of the RAG-based Prompt Reconstructor

This example demonstrates how to use the new RAG-based reconstruction method
as an alternative to the traditional content reference approach.
"""

from agentgraph.reconstruction import (
    reconstruct_prompts_from_knowledge_graph_rag,
    enrich_knowledge_graph_with_prompts_rag,
    RagPromptReconstructor
)

def example_usage():
    """Example of how to use the RAG-based prompt reconstructor."""
    
    # Sample knowledge graph data
    knowledge_graph = {
        "entities": [
            {
                "id": "user_001", 
                "name": "User inquiry about document loader",
                "type": "Input",
                "raw_prompt": "What is a document loader?"
            },
            {
                "id": "agent_001",
                "name": "Stereotypical Robot Named Robbie", 
                "type": "Agent",
                "raw_prompt": "You are a stereotypical robot named Robbie..."
            }
        ],
        "relations": [
            {
                "id": "rel_001",
                "source": "user_001",
                "target": "agent_001", 
                "type": "PERFORMS",
                "interaction_prompt": "What is a document loader?"
            }
        ]
    }
    
    # Sample original trace content
    original_trace = """
    User: What is a document loader?
    
    Agent: BEEP BOOP! Hello human! A document loader is a component in 
    LangChain that helps load documents from various sources. BEEP!
    
    There are many types like TextLoader, PDFLoader, CSVLoader, etc.
    Each one is designed for specific file formats. BOOP BEEP!
    
    Would you like me to explain any specific type? BEEP BOOP!
    """
    
    # Method 1: Using the pure function
    print("=== Using Pure Function ===")
    reconstructed_relations = reconstruct_prompts_from_knowledge_graph_rag(
        knowledge_graph=knowledge_graph,
        original_trace=original_trace,
        llm_config={"model": "gpt-5-mini", "temperature": 0.1}
    )
    
    for relation in reconstructed_relations:
        print(f"Relation: {relation['id']}")
        print(f"Type: {relation['type']}")
        print(f"Reconstructed Prompt:")
        print(relation['prompt'])
        print(f"Search Queries Used: {relation.get('search_queries_used', [])}")
        print("-" * 50)
    
    # Method 2: Using the class directly
    print("\n=== Using RagPromptReconstructor Class ===")
    reconstructor = RagPromptReconstructor(
        knowledge_graph=knowledge_graph,
        original_trace=original_trace,
        llm_config={"model": "gpt-5-mini", "temperature": 0.1}
    )
    
    # Reconstruct a specific relation
    specific_reconstruction = reconstructor.reconstruct_relation_prompt("rel_001")
    print(f"Specific Reconstruction:")
    print(f"Prompt: {specific_reconstruction['reconstructed_prompt']}")
    print(f"Method: {specific_reconstruction['reconstruction_method']}")
    
    # Method 3: Enrich entire knowledge graph
    print("\n=== Enriching Knowledge Graph ===")
    enriched_kg = enrich_knowledge_graph_with_prompts_rag(
        knowledge_graph=knowledge_graph,
        original_trace=original_trace
    )
    
    print(f"Original KG had {len(knowledge_graph.get('relations', []))} relations")
    print(f"Enriched KG has {len(enriched_kg.get('prompt_reconstructions', []))} reconstructed prompts")
    print(f"Reconstruction metadata: {enriched_kg.get('reconstruction_metadata', {})}")

if __name__ == "__main__":
    example_usage()