debug_vector_scores.py

import asyncio
import os
import sys
import numpy as np

# Add server directory to path
sys.path.append('/Users/mac/Projects/customerAgent/server')

from app.services.vector_db import VectorDB
from app.services.vector_operations import VectorOperations

async def debug_search():
    vdb = VectorDB()
    website_id = 19
    query = "what Healthcare Services do you provide"
    
    print(f"🚀 Debugging search for website {website_id}")
    print(f"🔍 Query: \"{query}\"")
    
    if not vdb.load(website_id):
        print(f"❌ Failed to load index for website {website_id}")
        return
        
    embedding = await VectorOperations.get_embedding(query, is_query=True)
    query_vector = np.array(embedding, dtype=np.float32)
    
    # Search WITHOUT thresholds
    results = vdb.search(query_vector, website_id, k=8, min_score=0.0)
    
    print(f"\n📊 Raw Results (min_score=0.0):")
    if not results:
        print("   No results found even with 0.0 threshold!")
    else:
        for i, (meta, score) in enumerate(results):
            text_preview = meta.get('text', '')[:60].replace('\n', ' ')
            print(f"   [{i}] Score: {score:.4f} | Truth: {meta.get('truth_level')} | Priority: {meta.get('priority')} | Text: {text_preview}...")

if __name__ == "__main__":
    asyncio.run(debug_search())