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.vectorstores import FAISS
from langchain.chains import RetrievalQA

# Google Generative AI integrations
from langchain_google_genai import GoogleGenerativeAI  # For LLM generation
from langchain_google_genai.embeddings import GoogleGenerativeAIEmbeddings  # For embeddings

# Load environment variables (GEMINI_API_KEY should be defined)
load_dotenv()
GEMINI_API_KEY = os.getenv("GEMINI_API_KEY")
if not GEMINI_API_KEY:
    raise ValueError("GEMINI_API_KEY not found in .env file")

# Configure the LLM using Google’s Gemini model.
# You can change the model name if needed (e.g., "gemini-pro", "gemini-1.5-flash-latest", etc.)
llm = GoogleGenerativeAI(model="gemini-1.5-flash-latest", google_api_key=GEMINI_API_KEY)

# Global variable for the RetrievalQA chain
qa_chain = None

@cl.on_chat_start
async def start_chat():
    """
    On chat start, this function loads a document from the provided URL using WebBaseLoader,
    splits it into chunks for retrieval, creates embeddings with Google’s embedding model,
    and builds a vector store (using FAISS). Finally, it creates a RetrievalQA chain that
    will retrieve relevant document sections and generate answers using the Gemini LLM.
    """
    global qa_chain

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

    # Split the document into chunks for effective retrieval
    text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
    docs = text_splitter.split_documents(documents)

    # Create embeddings using Google Generative AI embeddings
    embeddings = GoogleGenerativeAIEmbeddings(model="models/text-embedding-004", google_api_key=GEMINI_API_KEY)
    
    # Build a FAISS vector store for efficient similarity search
    vectorstore = FAISS.from_documents(docs, embeddings)
    retriever = vectorstore.as_retriever(search_kwargs={"k": 3})

    # Build the RetrievalQA chain that augments queries with the retrieved context
    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 questions about 'Künstliche Intelligenz'."
    ).send()

@cl.on_message
async def process_message(message: cl.Message):
    """
    When a user message arrives, this function uses the RetrievalQA chain to retrieve relevant
    context from the processed document, augment the user query, and generate an answer using
    the Gemini-based LLM.
    """
    global qa_chain
    if qa_chain is None:
        await cl.Message(content="❌ The document is still being loaded. Please wait a moment.").send()
        return

    # Retrieve user query and generate the answer using the chain
    query = message.content.strip()
    result = qa_chain.run(query)
    await cl.Message(content=result).send()