app.py

import streamlit as st
from langchain.embeddings.sentence_transformer import SentenceTransformerEmbeddings
from transformers import AutoModel
from pinecone import Pinecone
import os  # For environment variable access

# Replace with your Space's environment variable name for API key
API_KEY = os.environ.get("PINECONE_API_KEY")
#pc = Pinecone(api_key=PINECONE_API_KEY)
# Connect to Pinecone using the API key retrieved from the Space's environment variable
#client = IndexClient(api_key=API_KEY)
pc = Pinecone(api_key=API_KEY)

# Load pre-trained model (replace with your chosen model)
model =SentenceTransformerEmbeddings(model_name="all-MiniLM-L6-v2")

def process_and_search(query):
    # Preprocess user input (example: tokenization, normalization)
    #preprocessed_query = preprocess_query(query)  # Replace with your implementation

    # Encode the preprocessed query using the pre-trained model
    encoded_query = model.embed_query(query, return_tensors="pt")

    # Perform vector search in Pinecone
    results = pc.query(INDEX_NAME, encoded_query.cpu().numpy())

    # Process search results (example: extract answers, format display)
    processed_results = []
    for result in results:
        # Example processing: extract answer from metadata
        answer = result["metadata"]["answer"]  # Adapt based on your data structure
        processed_results.append(answer)

    return processed_results
st.title("Pinecone Search App")

user_query = st.text_area("Enter your question:", height=100)

if st.button("Search"):
    if user_query:
        # Process, search, and display results (call the process_and_search function)
        answers = process_and_search(user_query)
        st.write("Search Results:")
        for answer in answers:
            st.write(f"- {answer}")
    else:
        st.error("Please enter a question.")