app.py

import os
import gradio as gr
import fitz  # PyMuPDF
from langchain_community.vectorstores import FAISS
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_google_genai import ChatGoogleGenerativeAI
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser
from gtts import gTTS

# ----------------------------------------------------
# 1️⃣ Extract text from PDF
# ----------------------------------------------------
def extract_text_from_pdf(pdf_file):
    text = ""
    with fitz.open(stream=pdf_file.read(), filetype="pdf") as doc:
        for page in doc:
            text += page.get_text("text")
    return text

# ----------------------------------------------------
# 2️⃣ Create FAISS Vector Database
# ----------------------------------------------------
def create_vectorstore(text):
    text_splitter = RecursiveCharacterTextSplitter(chunk_size=800, chunk_overlap=100)
    docs = text_splitter.create_documents([text])
    embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2")
    vectorstore = FAISS.from_documents(docs, embeddings)
    return vectorstore

# ----------------------------------------------------
# 3️⃣ Initialize Google Gemini Model
# ----------------------------------------------------
def get_model():
    os.environ["GOOGLE_API_KEY"] = os.getenv("GOOGLE_API_KEY", "")
    return ChatGoogleGenerativeAI(model="gemini-pro", temperature=0.2)

# ----------------------------------------------------
# 4️⃣ Chatbot Logic
# ----------------------------------------------------
def chat_with_pdf(pdf_file, user_input, history):
    if pdf_file is None:
        return history + [["❌ Please upload a PDF file first.", ""]]

    # Extract and create FAISS
    pdf_text = extract_text_from_pdf(pdf_file)
    vectorstore = create_vectorstore(pdf_text)
    retriever = vectorstore.as_retriever(search_kwargs={"k": 3})

    # Retrieve relevant chunks
    docs = retriever.get_relevant_documents(user_input)
    context = "\n\n".join([d.page_content for d in docs])

    # Generate Answer
    prompt = ChatPromptTemplate.from_template(
        """
        You are a helpful Urdu assistant. Answer in Urdu (Roman Urdu is fine if needed).
        Context from PDF:
        {context}

        User Question:
        {question}

        Give your answer in Urdu (Roman Urdu allowed).
        """
    )

    llm = get_model()
    chain = prompt | llm | StrOutputParser()
    answer = chain.invoke({"context": context, "question": user_input})

    # Text-to-speech in Urdu
    tts = gTTS(answer, lang="ur")
    tts.save("response.mp3")

    history += [[user_input, answer]]
    return history, "response.mp3"

# ----------------------------------------------------
# 5️⃣ Gradio Interface
# ----------------------------------------------------
with gr.Blocks(title="📘 Urdu RAG Chatbot") as demo:
    gr.Markdown("## 🤖 Urdu RAG Chatbot — Ask questions from your PDF (Roman Urdu supported)")

    with gr.Row():
        pdf_file = gr.File(label="📄 Upload your PDF")
    
    chatbot = gr.Chatbot(label="Chat with your PDF (Urdu / Roman Urdu)")
    user_input = gr.Textbox(label="💬 Ask something about your PDF")
    audio_output = gr.Audio(label="🔊 Audio Answer", type="filepath")

    clear = gr.Button("🔄 Clear Chat")

    user_input.submit(chat_with_pdf, [pdf_file, user_input, chatbot], [chatbot, audio_output])
    clear.click(lambda: None, None, chatbot, queue=False)

demo.launch()