app.py

import os
import chainlit as cl
from dotenv import load_dotenv

# LangChain imports for retrieval and generation
from langchain.document_loaders import WebBaseLoader
from langchain.text_splitter import CharacterTextSplitter
from langchain.embeddings import OpenAIEmbeddings
from langchain.vectorstores import FAISS
from langchain.chains import RetrievalQA
from langchain.llms import OpenAI

# Load environment variables (e.g., OPENAI_API_KEY)
load_dotenv()

# Global variable to store our QA chain.
qa_chain = None

@cl.on_chat_start
async def start_chat():
    """
    When the chat starts, load the document using WebBaseLoader, split it into chunks,
    create embeddings, build a vector store, and finally initialize a RetrievalQA chain.
    This chain will serve as the backend for our RAG system.
    """
    global qa_chain

    # URL to crawl (German Wikipedia page on Künstliche Intelligenz)
    url = "https://de.wikipedia.org/wiki/K%C3%BCnstliche_Intelligenz"
    
    # Retrieve the document from the webpage
    loader = WebBaseLoader(url)
    documents = loader.load()  # returns a list of Document objects

    # Split the document into manageable chunks for better retrieval
    text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
    docs = text_splitter.split_documents(documents)
    
    # Create embeddings (make sure your OPENAI_API_KEY is set in your environment)
    embeddings = OpenAIEmbeddings()
    
    # Build a vector store from the documents using FAISS
    vectorstore = FAISS.from_documents(docs, embeddings)
    
    # Configure the retriever: retrieve the top 3 most relevant chunks
    retriever = vectorstore.as_retriever(search_kwargs={"k": 3})
    
    # Set up the language model (using OpenAI LLM here) with desired parameters
    llm = OpenAI(temperature=0)
    
    # Create a RetrievalQA chain that first retrieves relevant context and then generates an answer.
    qa_chain = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff", retriever=retriever)
    
    await cl.Message(
        content="✅ Document loaded and processed successfully! "
                "You can now ask me questions about 'Künstliche Intelligenz'."
    ).send()

@cl.on_message
async def process_question(message: cl.Message):
    """
    When a message is received, use the QA chain to process the query. The chain:
    1. Retrieves relevant document chunks.
    2. Augments your query with the retrieved context.
    3. Generates an answer via the language model.
    """
    global qa_chain

    if qa_chain is None:
        await cl.Message(content="❌ The document has not been loaded yet.").send()
        return

    # Get the user's query
    query = message.content.strip()
    
    # Process the query using the RetrievalQA chain
    result = qa_chain.run(query)
    
    # Send the answer back to the user
    await cl.Message(content=result).send()