app.py

import gradio as gr
import os
from langchain_community.document_loaders import PyPDFLoader
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_community.embeddings.fastembed import FastEmbedEmbeddings
from langchain_community.vectorstores import FAISS
from huggingface_hub import InferenceClient
from langchain_core.prompts import ChatPromptTemplate

# --- 1. Model Setup using HF Inference Client ---
HF_TOKEN = os.environ.get("HF_TOKEN", "")

if not HF_TOKEN:
    print("⚠️ Warning: HF_TOKEN not set. The app may not work properly.")

# Use InferenceClient directly instead of LangChain wrapper
client = InferenceClient(token=HF_TOKEN)

# --- 2. The Core Logic ---
def generate_question_paper(pdf_file, difficulty, num_questions):
    if not pdf_file:
        return "❌ Please upload a PDF file first."
    
    if not HF_TOKEN:
        return "❌ Error: HF_TOKEN not configured. Please add your Hugging Face token in Space Settings > Repository secrets."
    
    try:
        # A. Load PDF
        loader = PyPDFLoader(pdf_file.name)
        pages = loader.load()
        
        if not pages:
            return "❌ Error: Could not extract text from PDF. Please ensure it's a valid PDF with text content."
        
        # B. Split Text
        text_splitter = RecursiveCharacterTextSplitter(
            chunk_size=1000,
            chunk_overlap=100
        )
        chunks = text_splitter.split_documents(pages)
        
        # C. Vector Store (FAISS)
        embeddings = FastEmbedEmbeddings()
        vector_store = FAISS.from_documents(chunks, embeddings)
        
        # D. Retrieve Context
        retriever = vector_store.as_retriever(search_kwargs={"k": 7})
        context_docs = retriever.invoke("Key concepts and definitions")
        context_text = "\n\n".join([doc.page_content for doc in context_docs])
        
        # E. Create Prompt
        prompt = f"""You are an expert academic examiner. Create a formal Question Paper based ONLY on the context provided below.

CONTEXT:
{context_text}

INSTRUCTIONS:
- Difficulty: {difficulty}
- Total Questions: {num_questions}
- Format:
  Section A: Multiple Choice Questions (MCQs)
  Section B: Short Answer Questions
  Section C: Long Answer/Essay Questions
- Provide the Answer Key for MCQs at the very end.

Do not output conversational text. Output ONLY the exam paper in a well-formatted structure."""
        
        # F. Generate using Mistral (no license needed and works better)
        response = client.text_generation(
            prompt,
            model="mistralai/Mistral-7B-Instruct-v0.3",
            max_new_tokens=2000,
            temperature=0.7,
        )
        
        return response

    except Exception as e:
        return f"❌ Error: {str(e)}\n\nPlease check:\n1. PDF is valid and contains text\n2. HF_TOKEN is correctly set in Space secrets\n3. Try again or contact support"

# --- 3. The UI ---
with gr.Blocks(title="AI Question Paper Generator", theme=gr.themes.Soft(primary_hue="blue")) as demo:
    gr.Markdown("# 📄 AI Question Paper Generator")
    gr.Markdown("Powered by **Mistral 7B** via Hugging Face Inference API")
    gr.Markdown("⚡ Fast • 🎯 Accurate • 📚 Context-Aware")
    
    with gr.Row():
        with gr.Column(scale=1):
            pdf_input = gr.File(
                label="📄 Upload Study Material (PDF)",
                file_types=[".pdf"]
            )
            
            with gr.Group():
                difficulty = gr.Radio(
                    ["Easy", "Medium", "Hard"], 
                    label="🎚️ Difficulty Level", 
                    value="Medium"
                )
                num_questions = gr.Slider(
                    5, 20, value=10, step=1, 
                    label="📊 Total Questions"
                )
            
            btn = gr.Button("✨ Generate Question Paper", variant="primary", size="lg")
            
            gr.Markdown("""
            ### 📝 Instructions:
            1. Upload a PDF containing study material
            2. Select difficulty level
            3. Choose number of questions
            4. Click Generate!
            """)
        
        with gr.Column(scale=2):
            output = gr.Markdown(label="Generated Question Paper")

    btn.click(
        fn=generate_question_paper,
        inputs=[pdf_input, difficulty, num_questions],
        outputs=output
    )
    
    gr.Markdown("""
    ---
    **Note:** Set `HF_TOKEN` in your Space's Settings → Repository secrets.
    Get your token from https://huggingface.co/settings/tokens
    """)

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