app.py

import streamlit as st
import pinecone
import os  # To access environment variables
from sentence_transformers import SentenceTransformer
import numpy as np
from datasets import load_dataset

# Step 1: Get the Pinecone API key from the environment variable (Hugging Face secret)
pinecone_api_key = os.getenv('PINECONE_API_KEY')  # Fetch Pinecone API key from Hugging Face secrets

if not pinecone_api_key:
    st.error("Pinecone API key not found! Make sure to set the secret in Hugging Face settings.")
    st.stop()

# Initialize Pinecone client using the API key
pinecone.init(api_key=pinecone_api_key, environment="us-west1-gcp")  # Change the environment if needed

# Connect to your Pinecone index
index_name = "legal-docs-index-dji2ip8"  # Your Pinecone index name
index = pinecone.Index(index_name)

# Step 2: Load the sentence-transformers model for embeddings
model = SentenceTransformer("all-MiniLM-L6-v2")

# Step 3: Load dataset (for reference in your app)
dataset = load_dataset("macadeliccc/US-LegalKit", split="train")
law_texts = [item['text'] for item in dataset if 'text' in item]

# Step 4: Function to search Pinecone index
def search_pinecone(query, top_k=5):
    # Create an embedding for the user's query
    query_embedding = model.encode([query])
    
    # Query the Pinecone index for similar documents
    results = index.query(query_embedding, top_k=top_k, include_metadata=True)
    
    # Extract the text of the top-k results
    return [match['metadata']['text'] for match in results['matches']]

# Step 5: Streamlit UI
st.title("🔍 Legal AI Assistant (US-LegalKit)")

query = st.text_input("📌 Enter your legal query:")

if query:
    # Get the top results from Pinecone
    results = search_pinecone(query)
    
    st.write("### 📄 Relevant Legal Documents:")
    for i, doc in enumerate(results, 1):
        st.write(f"**{i}.** {doc[:500]}...")  # Show preview of the document