app.py

import streamlit as st
import asyncio
import os
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.prompts import SystemMessagePromptTemplate, HumanMessagePromptTemplate
from langchain.vectorstores import Chroma
from langchain.embeddings import OpenAIEmbeddings
from langchain.chat_models import ChatOpenAI
from PyPDF2 import PdfReader
import aiohttp
from io import BytesIO

# Set up API key
os.environ["OPENAI_API_KEY"] = st.secrets["OPENAI_API_KEY"]

# Set up prompts
system_template = "Use the following context to answer a user's question. If you cannot find the answer in the context, say you don't know the answer."
system_role_prompt = SystemMessagePromptTemplate.from_template(system_template)

user_prompt_template = "Context:\n{context}\n\nQuestion:\n{question}"
user_role_prompt = HumanMessagePromptTemplate.from_template(user_prompt_template)

# Define RetrievalAugmentedQAPipeline class
class RetrievalAugmentedQAPipeline:
    def __init__(self, llm: ChatOpenAI, vector_db: Chroma) -> None:
        self.llm = llm
        self.vector_db = vector_db

    async def arun_pipeline(self, user_query: str):
        context_docs = self.vector_db.similarity_search(user_query, k=2)
        context_list = [doc.page_content for doc in context_docs]
        context_prompt = "\n".join(context_list)
        
        max_context_length = 12000
        if len(context_prompt) > max_context_length:
            context_prompt = context_prompt[:max_context_length]
        
        formatted_system_prompt = system_role_prompt.format()
        formatted_user_prompt = user_role_prompt.format(question=user_query, context=context_prompt)

        response = await self.llm.agenerate([formatted_system_prompt, formatted_user_prompt])
        return {"response": response.generations[0][0].text, "context": context_list}

# PDF processing functions
async def fetch_pdf(session, url):
    async with session.get(url) as response:
        if response.status == 200:
            return await response.read()
        else:
            st.error(f"Failed to fetch PDF from {url}")
            return None

async def process_pdf(pdf_content):
    pdf_reader = PdfReader(BytesIO(pdf_content))
    text = "\n".join([page.extract_text() for page in pdf_reader.pages])
    text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
    return text_splitter.split_text(text)

@st.cache_resource
def initialize_pipeline():
    return asyncio.run(main())

# Main execution
async def main():
    pdf_urls = [
        "https://www.whitehouse.gov/wp-content/uploads/2022/10/Blueprint-for-an-AI-Bill-of-Rights.pdf",
        "https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.600-1.pdf",
    ]

    all_chunks = []
    async with aiohttp.ClientSession() as session:
        pdf_contents = await asyncio.gather(*[fetch_pdf(session, url) for url in pdf_urls])
        
    for pdf_content in pdf_contents:
        if pdf_content:
            chunks = await process_pdf(pdf_content)
            all_chunks.extend(chunks)

    st.write(f"Created {len(all_chunks)} chunks from {len(pdf_urls)} PDF files")

    embeddings = OpenAIEmbeddings()
    vector_db = Chroma.from_texts(all_chunks, embeddings)
    
    chat_openai = ChatOpenAI()
    return RetrievalAugmentedQAPipeline(vector_db=vector_db, llm=chat_openai)

# Streamlit UI
st.title("AI Bill of Rights Q&A")

pipeline = initialize_pipeline()

user_query = st.text_input("Enter your question about the AI Bill of Rights:")

if user_query:
    result = asyncio.run(pipeline.arun_pipeline(user_query))
    
    st.write("Response:")
    st.write(result["response"])
    
    st.write("Context used:")
    for i, context in enumerate(result["context"], 1):
        st.write(f"{i}. {context[:100]}...")

if __name__ == "__main__":
    st.run()