app.py

import os
import gradio as gr
from langchain_community.document_loaders import YoutubeLoader, PyPDFLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_huggingface import HuggingFaceEmbeddings, HuggingFaceEndpoint
from langchain_community.vectorstores import FAISS
from langchain.chains import RetrievalQA

# --- CONFIGURATION ---
hf_token = os.getenv("HF_TOKEN")
repo_id = "mistralai/Mistral-7B-Instruct-v0.3"

# --- HELPER: LLM SETUP ---
def get_llm():
    if not hf_token:
        raise ValueError("HF_TOKEN not found in secrets.")
    return HuggingFaceEndpoint(
        repo_id=repo_id,
        max_new_tokens=512,
        temperature=0.3,
        huggingfacehub_api_token=hf_token
    )

# --- STATE ---
vector_db_state = None

# --- 1. PROCESSING ENGINE ---
def process_content(url, file_obj):
    global vector_db_state
    msg = ""
    docs = []

    try:
        if url and "youtube.com" in url:
            loader = YoutubeLoader.from_youtube_url(url, add_video_info=True)
            docs = loader.load()
            msg = f"✅ Loaded YouTube: {docs[0].metadata['title']}"
        elif file_obj is not None:
            loader = PyPDFLoader(file_obj.name)
            docs = loader.load()
            msg = f"✅ Loaded PDF: {os.path.basename(file_obj.name)}"
        else:
            return "⚠️ Please provide a YouTube URL or upload a PDF.", None

        # Split text
        text_splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50)
        splits = text_splitter.split_documents(docs)

        # Create Vector Store
        embeddings = HuggingFaceEmbeddings(model_name="all-MiniLM-L6-v2")
        vector_db_state = FAISS.from_documents(splits, embeddings)
        
        return msg, vector_db_state

    except Exception as e:
        return f"❌ Error: {str(e)}", None

# --- 2. CORE FEATURES ---
def chat_engine(message, history):
    if vector_db_state is None:
        return "Please upload content first."
    
    try:
        llm = get_llm()
        qa_chain = RetrievalQA.from_chain_type(
            llm=llm,
            chain_type="stuff",
            retriever=vector_db_state.as_retriever()
        )
        response = qa_chain.invoke(message)
        return response['result']
    except Exception as e:
        return f"Error: {str(e)}"

def generate_summary():
    if vector_db_state is None:
        return "Please upload content first."
    
    try:
        llm = get_llm()
        retriever = vector_db_state.as_retriever(search_kwargs={"k": 5})
        docs = retriever.invoke("Summary")
        context = "\n\n".join([d.page_content for d in docs])
        
        prompt = f"""[INST] Summarize this content into bullet points:
        {context}
        [/INST]"""
        return llm.invoke(prompt)
    except Exception as e:
        return f"Error: {str(e)}"

def generate_quiz():
    if vector_db_state is None:
        return "Please upload content first."
    
    try:
        llm = get_llm()
        retriever = vector_db_state.as_retriever(search_kwargs={"k": 3})
        docs = retriever.invoke("Key Concepts")
        context = "\n\n".join([d.page_content for d in docs])
        
        prompt = f"""[INST] Create 3 multiple choice questions (with answers) based on this:
        {context}
        [/INST]"""
        return llm.invoke(prompt)
    except Exception as e:
        return f"Error: {str(e)}"

def generate_mindmap():
    if vector_db_state is None:
        return "Please upload content first."

    try:
        llm = get_llm()
        retriever = vector_db_state.as_retriever(search_kwargs={"k": 4})
        docs = retriever.invoke("Structure")
        context = "\n\n".join([d.page_content for d in docs])
        
        prompt = f"""[INST] Create a Mermaid.js mindmap (graph TD) code block based on this:
        {context}
        [/INST]"""
        return llm.invoke(prompt)
    except Exception as e:
        return f"Error: {str(e)}"

# --- 3. UI ---
with gr.Blocks(theme=gr.themes.Soft()) as demo:
    gr.Markdown("# 🧠 OpenLearn AI (Stable Build)")
    
    with gr.Row():
        with gr.Column(scale=1):
            yt_input = gr.Textbox(label="YouTube URL")
            pdf_input = gr.File(label="Upload PDF")
            process_btn = gr.Button("🚀 Process", variant="primary")
            status_output = gr.Textbox(label="Status", interactive=False)
            
            summ_btn = gr.Button("📝 Summary")
            quiz_btn = gr.Button("❓ Quiz")
            map_btn = gr.Button("🗺️ Mind Map")

        with gr.Column(scale=2):
            chatbot = gr.ChatInterface(fn=chat_engine)
            
    with gr.Row():
        summ_output = gr.Markdown(label="Summary")
        quiz_output = gr.Markdown(label="Quiz")
        map_output = gr.Code(label="Mind Map Code")

    process_btn.click(process_content, inputs=[yt_input, pdf_input], outputs=[status_output])
    summ_btn.click(generate_summary, inputs=None, outputs=summ_output)
    quiz_btn.click(generate_quiz, inputs=None, outputs=quiz_output)
    map_btn.click(generate_mindmap, inputs=None, outputs=map_output)

if __name__ == "__main__":
    demo.launch()