main.py

import json
import os
from typing import List, Dict, Any, Optional
from pydantic import BaseModel
import uvicorn
from fastapi import FastAPI, HTTPException
from pinecone import Pinecone , ServerlessSpec
import numpy as np
from openai import OpenAI

# Load environment variables
from dotenv import load_dotenv
load_dotenv()

# Get API keys from environment variables
PINECONE_API_KEY = os.getenv('PINECONE_API_KEY')
OPENAI_API_KEY = os.getenv('OPENAI_API_KEY')

if not PINECONE_API_KEY:
    raise ValueError("PINECONE_API_KEY environment variable not set")
if not OPENAI_API_KEY:
    raise ValueError("OPENAI_API_KEY environment variable not set")

# Create FastAPI app
app = FastAPI(title="E-Bikes Semantic Search API",
              description="API for finding similar e-bikes based on semantic search",
              version="1.0.0")

def build_filter(pt: Optional[str], cat: Optional[str]) -> dict | None:
    filt = {}
    if pt:
        filt["product_type"] = pt          # shorthand $eq
    if cat:
        filt["category"] = cat
    return filt or None

# Request and response models
class SearchRequest(BaseModel):
    description: str
    top_k: int = 3
    product_type: Optional[str] = None   # "ebike" or "escooter"
    category: Optional[str] = None       # e.g. "mountain"

class BikeMatch(BaseModel):
    id: str
    name: str
    type: str
    description: str
    score: float

class SearchResponse(BaseModel):
    matches: List[BikeMatch]

# Initialize OpenAI client
openai_client = OpenAI(api_key=OPENAI_API_KEY)

# Define the embedding model using OpenAI
class OpenAIEmbedder:
    def __init__(self, model_name="text-embedding-3-small"):
        self.model_name = model_name
        self.client = openai_client
        self.embedding_dimension = 1536  # Dimension of text-embedding-3-small

    def encode(self, texts):
        if isinstance(texts, str):
            texts = [texts]
        
        # Get embeddings from OpenAI
        response = self.client.embeddings.create(
            input=texts,
            model=self.model_name
        )
        
        # Extract embeddings from response
        embeddings = [item.embedding for item in response.data]
        return np.array(embeddings)

# Initialize Pinecone client
def initialize_pinecone():
    pc = Pinecone(api_key=PINECONE_API_KEY)
    
    # Define index name
    index_name = "ebikes-search"
    
    # Check if index already exists
    existing_indexes = pc.list_indexes().names()
    if index_name not in existing_indexes:
        # Create index with 1536 dimensions (matches text-embedding-3-small)
        pc.create_index(
            name=index_name,
            dimension=1536,
            metric="cosine",
            spec=ServerlessSpec(cloud="aws", region="us-east-1")
        )
        print(f"Created new index: {index_name}")
    
    # Connect to the index
    try:
        index = pc.Index(index_name)
        return index
    except Exception as e:
        print(f"Error connecting to Pinecone index: {e}")
        raise

# Load the e-bikes data
def load_ebikes_data(file_path="data.json"):
    try:
        with open(file_path, 'r') as f:
            data = json.load(f)
        return data.get('pogo-cycles-data', [])
    except Exception as e:
        print(f"Error loading e-bikes data: {e}")
        return []

# Create embeddings and upload to Pinecone
def create_and_upload_embeddings(ebikes_data, encoder, pinecone_index):
    # Prepare data for indexing
    ids = []
    descriptions = []
    metadata = []
    
    for bike in ebikes_data:
        ids.append(bike['id'])
        descriptions.append(bike['description'])
        metadata.append({
            'id': bike['id'],
            'name': bike['name'],
            'type': bike['product_type'],
            'description': bike['description'],
            'category': bike['category']
        })
    
    # Create embeddings
    embeddings = encoder.encode(descriptions)
    
    # Prepare vectors for Pinecone
    vectors_to_upsert = []
    for i in range(len(ids)):
        vector = {
            'id': ids[i],
            'values': embeddings[i].tolist(),
            'metadata': metadata[i]
        }
        vectors_to_upsert.append(vector)
    
    # Upsert vectors to Pinecone
    pinecone_index.upsert(vectors=vectors_to_upsert)
    print(f"Uploaded {len(vectors_to_upsert)} embeddings to Pinecone")

# Global variables for model and Pinecone index
encoder = None
pinecone_index = None

# Initialize data at startup
@app.on_event("startup")
async def startup_event():
    global encoder, pinecone_index
    
    print("Initializing OpenAI embedder...")
    encoder = OpenAIEmbedder()
    
    print("Connecting to Pinecone...")
    pinecone_index = initialize_pinecone()
    
    print("Loading e-bikes data...")
    ebikes_data = load_ebikes_data("data.json")
    
    if not ebikes_data:
        print("No e-bikes data found, skipping embedding creation")
        return
    
    print("Creating and uploading embeddings...")
    create_and_upload_embeddings(ebikes_data, encoder, pinecone_index)
    
    print("API startup completed successfully!")

@app.get("/health")
async def health_check():
    """Health check endpoint"""
    return {"status": "healthy"}

@app.post("/search", response_model=SearchResponse)
async def search_ebikes(request:SearchRequest):
    """
    Search for e-bikes similar to the provided description
    
    This endpoint uses semantic search to find e-bikes that match the user's description.
    """
    try:
        # Create embedding for the query
        query_embedding = encoder.encode(request.description)[0]
        filter_payload = build_filter(request.filters.get("product_type"), request.filters.get("category"))

        # Query Pinecone
        results = pinecone_index.query(
            vector=query_embedding.tolist(),
            top_k=3,
            include_metadata=True,
            filter=filter_payload
        )
        
        # Parse results
        matches = []
        for match in results.matches:
            bike_match = BikeMatch(
                id=match.metadata.get('id'),
                name=match.metadata.get('name'),
                type=match.metadata.get('type'),
                description=match.metadata.get('description'),
                score=float(match.score)
            )
            matches.append(bike_match)
        
        return SearchResponse(matches=matches)
    
    except Exception as e:
        print(f"Error during search: {e}")
        raise HTTPException(status_code=500, detail=f"Search failed: {str(e)}")

if __name__ == "__main__":
    uvicorn.run("main:app", host="0.0.0.0", port=8000, reload=True)