app.py

import gradio as gr
import os
import time
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 langchain_huggingface import HuggingFaceEndpoint
from langchain_core.prompts import ChatPromptTemplate

# --- 1. Model Setup ---
HF_TOKEN = os.environ.get("HF_TOKEN")

# Setup the Model (Llama 3 8B via API)
if HF_TOKEN:
    llm = HuggingFaceEndpoint(
        repo_id="meta-llama/Meta-Llama-3-8B-Instruct",
        temperature=0.5,
        max_new_tokens=4096,
        huggingfacehub_api_token=HF_TOKEN,
    )
else:
    llm = None

# --- 2. Logic ---
def generate_question_paper(pdf_file, difficulty, num_questions):
    # Error Handling for missing token
    if not HF_TOKEN:
        return "⚠️ Error: HF_TOKEN is missing. Please add it in Space Settings > Secrets.", "Error"
    
    if not pdf_file:
        return "⚠️ Please upload a PDF file first.", "Input Error"

    try:
        # Progress updates (simulated for UI feedback)
        yield "📂 Reading PDF...", "Processing"
        loader = PyPDFLoader(pdf_file.name)
        pages = loader.load()
        
        if not pages:
            return "❌ Error: The PDF appears to be empty or unreadable.", "Error"

        yield f"✂️ Splitting {len(pages)} pages...", "Processing"
        text_splitter = RecursiveCharacterTextSplitter(
            chunk_size=1000,
            chunk_overlap=100
        )
        chunks = text_splitter.split_documents(pages)

        yield "🧠 Analyzing content...", "Embedding"
        embeddings = FastEmbedEmbeddings()
        vector_store = FAISS.from_documents(chunks, embeddings)
        
        yield "🔍 Retrieving key concepts...", "Retrieving"
        retriever = vector_store.as_retriever(search_kwargs={"k": 5})
        context_docs = retriever.invoke("Summary of key topics, definitions, and important details")
        context_text = "\n\n".join([doc.page_content for doc in context_docs])

        # Prompt
        template = """
        You are an expert academic examiner. Create a rigorous Question Paper based ONLY on the provided context.
        
        CONTEXT:
        {context}
        
        INSTRUCTIONS:
        - Difficulty: {difficulty}
        - Total Questions: {num_questions}
        - Structure:
          * Part A: Multiple Choice ({num_questions} questions)
          * Part B: Short Answer (2 questions)
          * Part C: Essay/Long Answer (1 question)
        - Include an "Answer Key" section at the very bottom.
        
        OUTPUT FORMAT:
        Return valid Markdown. Use bold headers. Do not output conversational filler.
        """
        
        prompt = ChatPromptTemplate.from_template(template)
        chain = prompt | llm
        
        yield "✨ Generating final paper...", "Generating"
        response = chain.invoke({
            "context": context_text, 
            "difficulty": difficulty,
            "num_questions": num_questions
        })
        
        yield response, "Complete"

    except Exception as e:
        yield f"❌ System Error: {str(e)}", "Failed"

# --- 3. Custom UI ---
# Custom CSS for a professional look
custom_css = """
.container { max-width: 1200px; margin: auto; padding-top: 20px; }
.header-text { text-align: center; font-family: 'Helvetica', sans-serif; }
.header-text h1 { color: #2D3748; font-size: 3em; margin-bottom: 0px; }
.header-text h3 { color: #718096; font-weight: 300; }
.submit-btn { background: linear-gradient(90deg, #4F46E5 0%, #7C3AED 100%) !important; color: white !important; border: none !important; }
.status-bar { border: 1px solid #e2e8f0; background: #f7fafc; padding: 10px; border-radius: 8px; color: #4a5568; }
"""

theme = gr.themes.Soft(
    primary_hue="indigo",
    secondary_hue="blue",
    neutral_hue="slate",
    text_size="lg"
)

with gr.Blocks(theme=theme, css=custom_css, title="AI Exam Gen") as demo:
    
    with gr.Column(elem_classes="container"):
        # Header
        gr.HTML("""
            <div class="header-text">
                <h1>📝 AI Question Paper Generator</h1>
                <h3>Upload study material, get a formatted exam in seconds.</h3>
            </div>
        """)
        
        with gr.Row(variant="panel", equal_height=True):
            # Left Column: Inputs
            with gr.Column(scale=1):
                gr.Markdown("### 🛠️ Configuration")
                pdf_input = gr.File(
                    label="Upload PDF (Study Notes/Book)",
                    file_types=[".pdf"],
                    file_count="single",
                    height=100
                )
                
                with gr.Group():
                    difficulty = gr.Radio(
                        ["Easy", "Medium", "Hard"], 
                        label="Difficulty Level", 
                        value="Medium",
                        info="Adjusts complexity of questions."
                    )
                    
                    num_questions = gr.Slider(
                        minimum=5, 
                        maximum=20, 
                        value=10, 
                        step=1, 
                        label="Number of MCQs",
                        info="How many objective questions?"
                    )
                
                btn = gr.Button("✨ Generate Question Paper", elem_classes="submit-btn", variant="primary")
                status = gr.Textbox(label="Status", placeholder="Ready", interactive=False, max_lines=1)

            # Right Column: Output
            with gr.Column(scale=2):
                gr.Markdown("### 📄 Generated Exam")
                output = gr.Markdown(
                    label="Exam Paper", 
                    value="_Your generated question paper will appear here..._",
                    show_copy_button=True,
                    line_breaks=True
                )

    # Logic
    btn.click(
        fn=generate_question_paper,
        inputs=[pdf_input, difficulty, num_questions],
        outputs=[output, status]
    )

    # Footer
    gr.Markdown(
        """
        <div style="text-align: center; color: #a0aec0; margin-top: 40px;">
            Powered by Llama 3 • LangChain • Hugging Face
        </div>
        """
    )

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