import sys
import numpy as np
from sentence_transformers import SentenceTransformer

def main():
    print("=" * 60)
    print("LOCAL SEMANTIC CACHE SIMULATOR")
    print("=" * 60)
    
    # Initialize lightweight local embedding model
    print("Loading local embedding model (all-MiniLM-L6-v2)...")
    try:
        model = SentenceTransformer("all-MiniLM-L6-v2")
    except Exception as e:
        print(f"Error loading model: {e}")
        sys.exit(1)
        
    print("\nModel loaded successfully.")
    print("-" * 60)
    
    # Enforce strict threshold to avoid false-positives
    THRESHOLD = 0.92
    
    # User input
    sentence_1 = input("Enter Sentence 1 (The Baseline Cached Query):\n> ").strip()
    sentence_2 = input("\nEnter Sentence 2 (The New Incoming Query):\n> ").strip()
    
    if not sentence_1 or not sentence_2:
        print("\n[Error] Both sentences must contain text.")
        sys.exit(1)
        
    print("\nGenerating local embeddings and performing vector math...")
    
    # Generate embeddings (Returns 1D NumPy array for each string)
    emb1 = model.encode(sentence_1)
    emb2 = model.encode(sentence_2)
    
    # Calculate Cosine Similarity w NumPy vector operations
    dot_product = np.dot(emb1, emb2)
    norm_1 = np.linalg.norm(emb1)
    norm_2 = np.linalg.norm(emb2)
    
    similarity_score = dot_product / (norm_1 * norm_2)
    
    print("-" * 60)
    print(f"RESULTS:")
    print(f"-> Calculated Cosine Similarity: {similarity_score:.4f}")
    print(f"-> Target Safety Threshold:     {THRESHOLD:.4f}")
    
    # Evaluate Cache Efficacy
    if similarity_score >= THRESHOLD:
        print("\n[CACHE HIT] Returning cached response.")
    else:
        print("\n[CACHE MISS] Score below threshold. Request routed to LLM.")
    print("=" * 60)

if __name__ == "__main__":
    main()

# To do: see README.md