app.py

import gradio as gr
import os
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from accelerate import Accelerator
import re
import time
import PyPDF2
import sqlite3
import json # Import json for pretty printing parsed data

# --- Global Constants ---
MAX_QUIZ_QUESTIONS_UI = 20 # Define this globally so UI can access it
WEAK_THRESHOLD_PERCENTAGE = 65 # Percentage below which a topic is considered a weak area
quiz_context = {}
# --- Database Setup ---
DB_NAME = 'edututor.db'

def init_db():
    conn = sqlite3.connect(DB_NAME)
    cursor = conn.cursor()
    cursor.execute('''
        CREATE TABLE IF NOT EXISTS quiz_results (
            id INTEGER PRIMARY KEY AUTOINCREMENT,
            timestamp TEXT NOT NULL,
            topic TEXT NOT NULL,
            difficulty TEXT NOT NULL,
            score INTEGER,
            num_questions INTEGER
        )
    ''')
    conn.commit()
    conn.close()
    print(f"Database '{DB_NAME}' initialized successfully.")

def record_quiz_result(topic, difficulty, score, num_questions):
    conn = sqlite3.connect(DB_NAME)
    cursor = conn.cursor()
    timestamp = time.strftime('%Y-%m-%d %H:%M:%S')
    cursor.execute('''
        INSERT INTO quiz_results (timestamp, topic, difficulty, score, num_questions)
        VALUES (?, ?, ?, ?, ?)
    ''', (timestamp, topic, difficulty, score, num_questions))
    conn.commit()
    conn.close()
    print(f"Recorded quiz result: Topic='{topic}', Difficulty='{difficulty}', Score={score}/{num_questions}")

def get_performance_data():
    """Fetches raw performance data from the database."""
    conn = sqlite3.connect(DB_NAME)
    cursor = conn.cursor()
    cursor.execute('SELECT timestamp, topic, difficulty, score, num_questions FROM quiz_results ORDER BY timestamp DESC')
    data = cursor.fetchall()
    conn.close()
    return data

def get_performance_insights():
    """
    Analyzes quiz performance data and generates markdown for weak areas
    and suggested next steps, along with the dataframe for history.
    """
    performance_data = get_performance_data() # Fetch raw data

    # We return the actual data for the DataFrame, Gradio will handle updating its value
    df_data_for_update = performance_data

    if not performance_data:
        # Return empty data for DataFrame, and messages for markdown
        return [], "No quiz data available for analysis. Please take some quizzes first!", "No specific suggestions at this time."

    topic_scores = {} # {topic: [score_percentage, ...]}
    for timestamp, topic, difficulty, score, num_questions in performance_data:
        if num_questions > 0:
            percentage = (score / num_questions) * 100
            if topic not in topic_scores:
                topic_scores[topic] = []
            topic_scores[topic].append(percentage)

    weak_areas_list = []

    for topic, percentages in topic_scores.items():
        avg_percentage = sum(percentages) / len(percentages)
        num_quizzes = len(percentages)
        if avg_percentage < WEAK_THRESHOLD_PERCENTAGE:
            weak_areas_list.append({
                "topic": topic,
                "avg_score_percentage": avg_percentage,
                "num_quizzes": num_quizzes
            })

    # Sort weak areas by average score (lowest first)
    weak_areas_list.sort(key=lambda x: x['avg_score_percentage'])

    # --- Weak Areas Markdown ---
    weak_areas_markdown = ""
    if not weak_areas_list:
        weak_areas_markdown = "🎉 **Excellent work!** No significant weak areas detected based on your quiz performance. Keep up the great work!"
    else:
        weak_areas_markdown = "### Based on your performance, here are some areas where you might need more practice:\n\n"
        for area in weak_areas_list:
            weak_areas_markdown += f"- **{area['topic']}**: Average score of **{area['avg_score_percentage']:.1f}%** across {area['num_quizzes']} quiz(es).\n"

    # --- Suggested Steps Markdown ---
    suggested_steps_markdown = ""
    if weak_areas_list:
        suggested_steps_markdown += "### Suggested Next Steps:\n\n"
        suggested_steps_markdown += "🌟 **Focus on these topics:**\n"
        for area in weak_areas_list:
            suggested_steps_markdown += f"- Try generating a new quiz specifically on **{area['topic']}** (e.g., using the 'Generate Quiz' tab).\n"
        suggested_steps_markdown += "\n💡 Practice makes perfect! Revisit relevant materials and re-take quizzes on these subjects."
    else:
        suggested_steps_markdown = "### Suggested Next Steps:\n\n"
        suggested_steps_markdown += "Keep exploring new topics or challenge yourself with a 'hard' difficulty quiz! You're doing great!"

    return df_data_for_update, weak_areas_markdown, suggested_steps_markdown


# --- Model Loading (Local Inference) ---
GRANITE_MODEL_NAME = "ibm-granite/granite-3.3-2b-instruct"

tokenizer = None
model = None
device = "cuda" if torch.cuda.is_available() else "cpu"

def load_granite_model():
    global tokenizer, model
    if model is not None and tokenizer is not None:
        return tokenizer, model

    print(f"Loading {GRANITE_MODEL_NAME} on {device}...")
    try:
        tokenizer = AutoTokenizer.from_pretrained(GRANITE_MODEL_NAME)
        model = AutoModelForCausalLM.from_pretrained(
            GRANITE_MODEL_NAME,
            device_map="auto",
            torch_dtype=torch.float16
        )
        model.eval()
        print(f"Successfully loaded model: {GRANITE_MODEL_NAME} locally!")
        print(f"Model is running on: {device}")
        return tokenizer, model
    except Exception as e:
        print(f"Error loading model {GRANITE_MODEL_NAME}: {e}")
        print("Possible reasons: Insufficient GPU RAM, model not found, or network issues during download.")
        print("Consider trying a smaller model like 'ibm-granite/granite-3.0-2b-instruct' or upgrading Colab runtime.")
        return None, None

# Try loading the actual model, if it fails, fallback to mock objects.
try:
    tokenizer, model = load_granite_model()
    if tokenizer is None or model is None:
        raise Exception("Model or tokenizer failed to load, falling back to mock.")
except Exception as e:
    print(f"INFO: Failed to load actual model ({e}). Using mock model for demonstration.")
    class MockModel:
        def generate(self, input_ids, max_new_tokens, temperature, top_p, do_sample, pad_token_id, eos_token_id):
            text = self.tokenizer.decode(input_ids[0], skip_special_tokens=True)
            if "summarize" in text.lower():
                return self.tokenizer.encode("This is a mock summary of your text.")
            elif "refine" in text.lower():
                return self.tokenizer.encode("This is your mock refined text.")
            elif "meaning of" in text.lower():
                return self.tokenizer.encode("Mock meaning: A mock object is a simulated object.")
            elif "translate" in text.lower():
                return self.tokenizer.encode("Mock translation.")
            elif "quiz" in text.lower():
                mock_quiz_output = """
1. Question: What is the capital of France?
    A. Berlin
    B. Madrid
    C. Paris
    D. Rome

    Correct Answer: C
    Explanation: Paris is the capital and most populous city of France.

2. Question: Which planet is known as the Red Planet?
    A. Earth
    B. Mars
    C. Jupiter
    D. Venus

    Correct Answer: B
    Explanation: Mars is often referred to as the Red Planet because of its reddish appearance.

3. Question: What is the largest ocean on Earth?
    A. Atlantic Ocean
    B. Indian Ocean
    C. Arctic Ocean
    D. Pacific Ocean

    Correct Answer: D
    Explanation: The Pacific Ocean is the largest and deepest of Earth's five oceanic divisions.
"""
                return self.tokenizer.encode(mock_quiz_output)
            return self.tokenizer.encode("Hello! This is a mock response from Edu-Tutor AI.")

    tokenizer = AutoTokenizer.from_pretrained("gpt2") # Using a small, easily downloadable tokenizer for mock
    model = MockModel()
    model.tokenizer = tokenizer # Attach tokenizer to mock model for encoding/decoding

# Ensure accelerator is prepared even for mock model (if using torch operations)
accelerator = Accelerator()
# Note: For the MockModel, accelerator.prepare might not be strictly necessary if it doesn't use torch tensors internally
# but keeping it for consistency if you swap to a real model.
# If you get errors here with the mock model, you can remove this line for the mock setup.
# model, tokenizer = accelerator.prepare(model, tokenizer)


# --- Helper Functions for Model Inference ---

STREAM_DELAY = 0.005 # seconds per word

def generate_response(prompt, chat_history_for_model):
    if model is None or tokenizer is None:
        yield "Error: Model not loaded. Please check Colab runtime and GPU memory."
        return

    messages = []
    for human, ai in chat_history_for_model:
        messages.append({"role": "user", "content": human if human is not None else ""})
        messages.append({"role": "assistant", "content": ai if ai is not None else ""})
    messages.append({"role": "user", "content": prompt})

    input_text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
    input_ids = tokenizer(input_text, return_tensors="pt").to(device)

    current_response = ""
    try:
        input_token_len = input_ids.input_ids.shape[1]

        # Ensure pad_token_id and eos_token_id are set for generate
        if tokenizer.pad_token_id is None:
            tokenizer.pad_token_id = tokenizer.eos_token_id # Often EOS token is used as pad_token_id
        if tokenizer.eos_token_id is None: # Fallback if tokenizer doesn't provide one
             # For some models/tokenizers, you might manually set an arbitrary ID or handle it differently
             # For gpt2, eos_token_id is typically 50256
            if hasattr(tokenizer, 'default_end_token_id'):
                tokenizer.eos_token_id = tokenizer.default_end_token_id
            elif tokenizer.vocab_size > 0:
                tokenizer.eos_token_id = tokenizer.vocab_size - 1 # Last token ID as a fallback


        full_output_tokens = model.generate(
            **input_ids,
            max_new_tokens=100,
            temperature=0.7,
            do_sample=True,
            pad_token_id=tokenizer.pad_token_id,
            eos_token_id=tokenizer.eos_token_id
        )
        full_response = tokenizer.decode(full_output_tokens[0][input_token_len:], skip_special_tokens=True).strip()

        words = full_response.split(" ")
        for i in range(len(words)):
            current_response += words[i] + " "
            yield current_response
            time.sleep(STREAM_DELAY)

        return

    except Exception as e:
        yield f"Error during inference: {e}"

# Function to read text from PDF
def read_pdf_text(pdf_file_path):
    text = ""
    if pdf_file_path is None:
        return ""
    try:
        with open(pdf_file_path, 'rb') as file:
            reader = PyPDF2.PdfReader(file)
            for page_num in range(len(reader.pages)):
                page = reader.pages[page_num]
                text += page.extract_text()
        return text
    except Exception as e:
        return f"Error reading PDF: {e}"

# --- Quiz Parsing Function ---
def parse_quiz_text(raw_quiz_text):
    """
    Parses the raw quiz text generated by the LLM into a structured list of dictionaries.
    Each dictionary represents a question.
    """
    parsed_questions = []

    # Regex to find each question block
    # It looks for:
    # 1. An optional number (e.g., "1.")
    # 2. " Question:" followed by the question text
    # 3. Lines starting with A., B., C., D. for options
    # 4. "Correct Answer: [Letter]"
    # 5. "Explanation: [Text]"
    question_pattern = re.compile(
        r'(\d*\.?\s*Q(?:uestion)?:\s*(.*?)\n' # Optional number, "Question:", then question text (group 2)
        r'\s*A\.\s*(.*?)\n' # Option A (group 3)
        r'\s*B\.\s*(.*?)\n' # Option B (group 4)
        r'\s*C\.\s*(.*?)\n' # Option C (group 5)
        r'\s*D\.\s*(.*?)\n' # Option D (group 6)
        r'\s*Correct Answer:\s*([A-D])\s*\n' # Correct Answer letter (group 7)
        r'\s*Explanation:\s*(.*?)(?=\n\d*\.?\s*Q(?:uestion)?:|\Z))', # Explanation (group 8), lookahead for next question or end of string
        re.DOTALL # . matches newlines
    )

    matches = question_pattern.findall(raw_quiz_text)

    for match in matches:
        question_text = match[1].strip()
        options = {
            'A': match[2].strip(),
            'B': match[3].strip(),
            'C': match[4].strip(),
            'D': match[5].strip()
        }
        correct_answer = match[6].strip()
        explanation = match[7].strip()

        # Clean up question text if it has extra "Question:" or leading/trailing spaces
        # Also remove the leading number if present, as it's just for display initially
        question_text = re.sub(r'^\d+\.\s*Question:\s*', '', question_text, flags=re.IGNORECASE).strip()

        parsed_questions.append({
            'question_text': question_text,
            'options': options,
            'correct_answer': correct_answer,
            'explanation': explanation
        })

    print(f"DEBUG: Parsed {len(parsed_questions)} questions.")
    return parsed_questions


# --- Quiz Generation and Scoring Functions ---

quiz_context = {
    "topic": "",
    "difficulty": "",
    "parsed_quiz": []
}


def generate_quiz_for_display(quiz_topic, num_quiz_questions_str, difficulty="medium", pdf_file=None):
    # Initialize a list of gr.update objects for all expected outputs
    # The first element is for parsed_quiz_data_state, and should be gr.update()
    updates = [
        gr.update(), # 0: parsed_quiz_data_state (will be updated with actual data) - CORRECTED
        gr.update(visible=False, value=""), # 1: raw_quiz_output_display (clear and hide)
        gr.update(visible=False), # 2: submit_quiz_button
        gr.update(value="", visible=False), # 3: quiz_results_display
        gr.update(visible=False), # 4: start_new_quiz_button
        gr.update(visible=False) # 5: interactive_quiz_column - initially hidden for reset
    ]
    # Add updates for all possible question/radio pairs, initially hidden
    for i in range(MAX_QUIZ_QUESTIONS_UI):
        updates.append(gr.update(value="", visible=False)) # Q_md
        updates.append(gr.update(choices=[], value=None, visible=False)) # Q_radio

    # Yield initial reset state (important for clearing previous quiz display)
    yield tuple(updates)

    if model is None or tokenizer is None:
        print(f"DEBUG: Model or tokenizer not loaded at start of generate_quiz_for_display. Current time: {time.time()}")
        updates[1] = gr.update(value="Error: Model not loaded. Please check Colab runtime and GPU memory.", visible=True)
        yield tuple(updates)
        return

    try:
        num_quiz_questions = int(num_quiz_questions_str)
        if not (1 <= num_quiz_questions <= MAX_QUIZ_QUESTIONS_UI):
            raise ValueError(f"Number of questions must be between 1 and {MAX_QUIZ_QUESTIONS_UI}.")
    except ValueError as e:
        print(f"DEBUG: Invalid number of questions: {e}. Current time: {time.time()}")
        updates[1] = gr.update(value=f"Invalid number of questions: {e}. Please enter a whole number between 1 and {MAX_QUIZ_QUESTIONS_UI}.", visible=True)
        yield tuple(updates)
        return

    context_text = ""
    effective_topic = ""
    if pdf_file:
        context_text = read_pdf_text(pdf_file.name)
        if "Error reading PDF" in context_text:
            print(f"DEBUG: Error reading PDF: {context_text}. Current time: {time.time()}")
            updates[1] = gr.update(value=context_text, visible=True)
            yield tuple(updates)
            return
        if not context_text.strip():
            print(f"DEBUG: PDF file is empty or could not be read (empty context_text). Current time: {time.time()}")
            updates[1] = gr.update(value="PDF file is empty or could not be read.", visible=True)
            yield tuple(updates)
            return
        effective_topic = f"PDF: {os.path.basename(pdf_file.name)}"
    elif quiz_topic and quiz_topic.strip():
        context_text = quiz_topic
        effective_topic = quiz_topic
    else:
        print(f"DEBUG: No topic or PDF provided. Current time: {time.time()}")
        updates[1] = gr.update(value="Please provide a topic for the quiz or upload a PDF.", visible=True)
        yield tuple(updates)
        return

    if not context_text.strip():
        print(f"DEBUG: Final context_text is empty after processing. Current time: {time.time()}")
        updates[1] = gr.update(value="A valid topic or non-empty PDF content is required to generate a quiz.", visible=True)
        yield tuple(updates)
        return

    difficulty_instruction = ""
    if difficulty == "easy":
        difficulty_instruction = "Make the questions relatively easy and straightforward."
    elif difficulty == "hard":
        difficulty_instruction = "Make the questions challenging and detailed, requiring deeper understanding."
    else: # medium
        difficulty_instruction = "Make the questions of medium difficulty."

    quiz_prompt = f"""Generate {num_quiz_questions} multiple-choice questions about the following text or topic:
    {context_text}

    {difficulty_instruction}
    For each question, provide exactly 4 options (A, B, C, D), clearly state the single correct answer letter, and give an **EXTREMELY brief, 1-sentence explanation. Be very concise.**
    Focus on **very short questions** and **short options**. No introductory or concluding remarks. Just the questions.
    **IMPORTANT: Prepend each "Question:" with its number (e.g., "1. Question:", "2. Question:").**
    Format each question STRICTLY like this, with no extra text or numbering outside this pattern:

    [Number]. Question: [Short question text]
    A. [Option A]
    B. [Option B]
    C. [Option C]
    D. [Option D]

    Correct Answer: [Letter]
    Explanation: [Very brief explanation, 1 sentence max, no fluff]

    Example:
    1. Question: Capital of France?
    A. Berlin
    B. Madrid
    C. Paris
    D. Rome

    Correct Answer: C
    Explanation: Paris is the capital.
    """
    print(f"DEBUG: Quiz prompt length: {len(quiz_prompt)}. First 200 chars: {quiz_prompt[:200]}. Current time: {time.time()}")

    messages = [{"role": "user", "content": quiz_prompt}]
    input_text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)

    if not input_text.strip():
        print(f"DEBUG: Tokenizer applied chat template resulted in empty input_text. Current time: {time.time()}")
        updates[1] = gr.update(value="Error: Could not create a valid prompt for the quiz. This might happen if the topic or PDF content is too short or invalid.", visible=True)
        yield tuple(updates)
        return

    quiz_input_ids = None
    try:
        quiz_input_ids = tokenizer(input_text, return_tensors="pt").to(device)
        print(f"DEBUG: Tokenized input_ids shape: {quiz_input_ids.input_ids.shape}. Current time: {time.time()}")
    except Exception as e:
        print(f"DEBUG: Error tokenizing input for quiz: {e}. Current time: {time.time()}")
        updates[1] = gr.update(value=f"Error tokenizing input for quiz: {e}. Please ensure input text is valid.", visible=True)
        yield tuple(updates)
        return

    try:
        MAX_TOKENS_PER_QUIZ_QUESTION = 70
        quiz_max_new_tokens = num_quiz_questions * MAX_TOKENS_PER_QUIZ_QUESTION
        print(f"DEBUG: Generating with max_new_tokens={quiz_max_new_tokens}. Current time: {time.time()}")

        full_output_tokens = model.generate(
            **quiz_input_ids,
            max_new_tokens=quiz_max_new_tokens,
            temperature=0.7,
            do_sample=True,
            pad_token_id=tokenizer.eos_token_id
        )
        raw_quiz_text = tokenizer.decode(full_output_tokens[0][quiz_input_ids.input_ids.shape[1]:], skip_special_tokens=True)

        # --- Temporarily show raw quiz text for debugging ---
        # You can set visible=False after testing
        updates[1] = gr.update(value=raw_quiz_text, visible=True, label="Raw Quiz Output (for debugging)")
        yield tuple(updates) # Yield this update so you can see the raw output while parsing happens

        print(f"DEBUG: Raw quiz text generated. Length: {len(raw_quiz_text)}. First 500 chars:\n{raw_quiz_text[:500]}. Current time: {time.time()}")

        parsed_quiz_data = parse_quiz_text(raw_quiz_text)
        print("DEBUG: Parsed quiz data:")
        print(json.dumps(parsed_quiz_data, indent=2))
        print(f"DEBUG: Number of questions parsed: {len(parsed_quiz_data)}. Current time: {time.time()}")

        if not parsed_quiz_data:
            print(f"DEBUG: No questions parsed from the generated text. Current time: {time.time()}")
            updates[1] = gr.update(value="Could not generate a valid quiz with the requested format. Please try again with a different topic or fewer questions.", visible=True)
            yield tuple(updates)
            return

        # Store for later use by scoring function
        quiz_context["topic"] = effective_topic
        quiz_context["difficulty"] = difficulty
        quiz_context["parsed_quiz"] = parsed_quiz_data

        # Update the initial components
        updates[0] = gr.update(value=parsed_quiz_data) # Update gr.State with the parsed data
        print(f"DEBUG: Value for parsed_quiz_data_state update: {len(parsed_quiz_data)} questions.")

        updates[1] = gr.update(visible=False, value="") # Hide the raw text output after successful parsing
        updates[2] = gr.update(visible=True) # Show the submit button
        updates[3] = gr.update(value="", visible=False) # Clear and hide previous results
        updates[4] = gr.update(visible=False) # Hide start new quiz button initially
        updates[5] = gr.update(visible=True) # Make the interactive quiz column visible here

        # Populate quiz questions and options.
        output_idx = 6 # Start index for question/radio pairs after the initial 6 common components
        for i, q in enumerate(parsed_quiz_data):
            if i < MAX_QUIZ_QUESTIONS_UI: # Ensure we don't exceed placeholder count
                question_label = f"{i+1}. {q['question_text']}"
                choices = [f"A. {q['options']['A']}", f"B. {q['options']['B']}", f"C. {q['options']['C']}", f"D. {q['options']['D']}"]

                # Update question Markdown and Radio button using gr.update
                updates[output_idx] = gr.update(value=question_label, visible=True)
                updates[output_idx + 1] = gr.update(choices=choices, value=None, label=f"Select your answer for Q{i+1}", visible=True)
            else:
                # This should ideally not happen if num_quiz_questions <= MAX_QUIZ_QUESTIONS_UI
                # But for safety, ensure any remaining placeholders are hidden.
                updates[output_idx] = gr.update(value="", visible=False)
                updates[output_idx + 1] = gr.update(value=None, visible=False, choices=[])
            output_idx += 2

        # Hide any remaining unused placeholder quiz elements
        for i in range(len(parsed_quiz_data), MAX_QUIZ_QUESTIONS_UI):
            updates[output_idx] = gr.update(value="", visible=False)
            updates[output_idx + 1] = gr.update(value=None, visible=False, choices=[])
            output_idx += 2

        print(f"DEBUG: Gradio UI updates prepared for interactive quiz. Current time: {time.time()}")
        yield tuple(updates) # Yield the populated quiz UI

    except Exception as e:
        print(f"DEBUG: Error during quiz generation inference or parsing: {e}. Current time: {time.time()}")
        updates[1] = gr.update(value=f"Error generating or parsing quiz: {e}. Please check Colab console for details.", visible=True)
        yield tuple(updates)


# Make sure quiz_context is defined globally at the top of your script, e.g.:
# quiz_context = {}

def submit_quiz(*user_answers_raw): # Removed parsed_quiz_data as an input argument
    """
    Scores the quiz based on user answers and provides detailed feedback.
    *user_answers_raw will be a tuple where each element is the selected option string for a question.
    """
    # --- DEBUGGING PRINTS: Access quiz_data from global quiz_context ---
    print(f"DEBUG: submit_quiz called.")

    # Get parsed quiz data directly from the global quiz_context
    parsed_quiz_data = quiz_context.get("parsed_quiz", [])

    print(f"DEBUG: Type of parsed_quiz_data (from quiz_context): {type(parsed_quiz_data)}")
    print(f"DEBUG: Content of parsed_quiz_data (from quiz_context): {parsed_quiz_data}")
    print(f"DEBUG: Length of parsed_quiz_data (from quiz_context): {len(parsed_quiz_data) if parsed_quiz_data is not None else 'None'}")
    # --- END DEBUGGING PRINTS ---

    # Prepare outputs list, matching the order of submit_quiz_outputs
    outputs = []

    # First, handle the common UI elements:
    outputs.append(gr.update(visible=False)) # submit_quiz_button (will be hidden)
    outputs.append(gr.update(value="", visible=False)) # quiz_results_display (cleared and hidden, will be updated)
    outputs.append(gr.update(visible=True)) # start_new_quiz_button (will be updated) - Always show this after submit
    outputs.append(gr.update(visible=False)) # interactive_quiz_column (will be hidden)

    # Initialize updates for all quiz question and radio button components
    # These must be included in the outputs tuple, even if they are hidden
    for _ in range(MAX_QUIZ_QUESTIONS_UI):
        outputs.append(gr.update(value="", visible=False)) # quiz_question_md[i] - No 'choices' here
        outputs.append(gr.update(value=None, visible=False, choices=[])) # quiz_options_radio[i] - important to clear choices too

    # --- Now check if the list retrieved from quiz_context is empty ---
    if not parsed_quiz_data: # Check if the list 'parsed_quiz_data' is empty
        print("DEBUG: parsed_quiz_data (from quiz_context) is empty. Returning 'No quiz data found' message.")
        outputs[1] = gr.update(value="No quiz data found. Please generate a quiz first.", visible=True)
        outputs[2] = gr.update(visible=True) # Show start new quiz button
        return tuple(outputs)

    score = 0
    feedback_markdown = "## Quiz Results\n\n"

    for i, question in enumerate(parsed_quiz_data): # Iterate directly over the list
        user_answer_display = "No answer provided"
        user_choice_letter = None

        if i < len(user_answers_raw) and user_answers_raw[i] is not None:
            user_answer_display = user_answers_raw[i]
            # Extract just the letter (A, B, C, D) from user_answer_raw (e.g., "A. Option Text")
            if isinstance(user_answers_raw[i], str) and len(user_answers_raw[i]) >= 1:
                user_choice_letter = user_answers_raw[i][0]

        correct_answer_letter = question['correct_answer']

        is_correct = (user_choice_letter == correct_answer_letter)
        if is_correct:
            score += 1
            feedback_markdown += f"✅ **Question {i+1}: Correct!**\n"
        else:
            feedback_markdown += f"❌ **Question {i+1}: Incorrect.**\n"

        feedback_markdown += f"**Your Answer:** {user_answer_display}\n"
        feedback_markdown += f"**Correct Answer:** {correct_answer_letter}. {question['options'].get(correct_answer_letter, 'Option not found')}\n"
        feedback_markdown += f"**Explanation:** {question['explanation']}\n\n---\n\n"

    total_questions = len(parsed_quiz_data) # Get length directly from the list
    final_score_text = f"## Your Score: {score} out of {total_questions}\n\n"
    feedback_markdown = final_score_text + feedback_markdown

    # Record result to database - ensure score and total_questions are integers
    record_quiz_result(quiz_context["topic"], quiz_context["difficulty"], int(score), int(total_questions))

    # Update the relevant output components in the 'outputs' list
    outputs[0] = gr.update(visible=False) # submit_quiz_button
    outputs[1] = gr.update(value=feedback_markdown, visible=True) # quiz_results_display
    outputs[2] = gr.update(visible=True) # start_new_quiz_button
    outputs[3] = gr.update(visible=False) # interactive_quiz_column (hide this after submission)

    # All other question/radio updates are already set to hidden/cleared by the initial `outputs` list setup.

    return tuple(outputs)
def start_new_quiz_ui_reset():
    """Resets the quiz UI to its initial state, hiding quiz and results elements."""
    outputs = [
        gr.update(value="", visible=True), # quiz_topic_input
        gr.update(value="3", visible=True), # num_questions_input
        gr.update(value="medium", visible=True), # quiz_difficulty
        gr.update(value=None, visible=True), # pdf_upload_input
        gr.update(visible=True), # generate_quiz_button
        gr.update(value="", visible=False), # raw_quiz_output_display (ensure hidden, now a Textbox)
        gr.update(visible=False), # submit_quiz_button (hidden)
        gr.update(value="", visible=False), # quiz_results_display (hidden)
        gr.update(visible=False), # start_new_quiz_button (hidden)
        gr.update(visible=False) # interactive_quiz_column (hide this too)
    ]
    # Hide all question fields and reset their values
    for i in range(MAX_QUIZ_QUESTIONS_UI):
        outputs.extend([
            gr.update(value="", visible=False), # quiz_question_md[i] - No 'choices' here
            gr.update(value=None, visible=False, choices=[]) # quiz_options_radio[i] - important to clear choices too
        ])

    return tuple(outputs)


# --- Summarizer, Refiner, Word, Translate Functions (Unchanged - ensure they use gr.update for outputs) ---
def summarize_text(input_text, summary_length="short"):
    if model is None or tokenizer is None:
        yield gr.update(value="Error: Model not loaded. Please check Colab runtime and GPU memory.")
        return
    if not input_text:
        yield gr.update(value="Please provide text to summarize.")
        return

    if summary_length == "short":
        prompt_instruction = "Summarize the following text concisely:"
        max_output_tokens = 100
    elif summary_length == "detailed":
        prompt_instruction = "Provide a detailed summary of the following text:"
        max_output_tokens = 250
    else: # simple_explanation
        prompt_instruction = "Explain the following text in simple terms, suitable for a beginner:"
        max_output_tokens = 200

    summary_prompt = f"{prompt_instruction}\n\n{input_text}"

    messages = [{"role": "user", "content": summary_prompt}]
    input_text_tokenized = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
    summary_input_ids = tokenizer(input_text_tokenized, return_tensors="pt").to(device)

    current_summary = ""
    try:
        full_output_tokens = model.generate(
            **summary_input_ids,
            max_new_tokens=max_output_tokens,
            temperature=0.7,
            do_sample=True,
            pad_token_id=tokenizer.eos_token_id
        )
        summary_raw = tokenizer.decode(full_output_tokens[0][summary_input_ids.input_ids.shape[1]:], skip_special_tokens=True)

        words = summary_raw.split(" ")
        for i in range(len(words)):
            current_summary += words[i] + " "
            yield gr.update(value=current_summary) # Ensure outputs are gr.update
            time.sleep(STREAM_DELAY)

        return

    except Exception as e:
        yield gr.update(value=f"Error during summarization: {e}") # Ensure outputs are gr.update


def refine_text(input_text):
    if model is None or tokenizer is None:
        yield gr.update(value="Error: Model not loaded. Please check Colab runtime and GPU memory.")
        return
    if not input_text:
        yield gr.update(value="Please provide text to refine.")
        return

    refine_prompt = f"""Review the following text for grammar, spelling, punctuation, and clarity. Provide corrected sentences and suggest improvements for style and conciseness. Explain any major changes.

    Original Text:
    {input_text}

    Refined Text:
    """

    messages = [{"role": "user", "content": refine_prompt}]
    input_text_tokenized = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
    refine_input_ids = tokenizer(input_text_tokenized, return_tensors="pt").to(device)

    current_refined_text = ""
    try:
        full_output_tokens = model.generate(
            **refine_input_ids,
            max_new_tokens=400,
            temperature=0.7,
            do_sample=True,
            pad_token_id=tokenizer.eos_token_id
        )
        refined_text_raw = tokenizer.decode(full_output_tokens[0][refine_input_ids.input_ids.shape[1]:], skip_special_tokens=True)

        words = refined_text_raw.split(" ")
        for i in range(len(words)):
            current_refined_text += words[i] + " "
            yield gr.update(value=current_refined_text) # Ensure outputs are gr.update
            time.sleep(STREAM_DELAY)

        return

    except Exception as e:
        yield gr.update(value=f"Error refining text: {e}") # Ensure outputs are gr.update

# Function for Word Meaning & Usage
def get_word_meaning_and_usage(word):
    if model is None or tokenizer is None:
        yield gr.update(value="Error: Model not loaded.")
        return
    if not word:
        yield gr.update(value="Please enter a word.")
        return

    prompt = f"""Provide the meaning of the word '{word}' and demonstrate its usage in two different example sentences.
    Format your response clearly.

    Word: {word}
    Meaning:
    Example 1:
    Example 2:
    """
    messages = [{"role": "user", "content": prompt}]
    input_text_tokenized = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
    input_ids = tokenizer(input_text_tokenized, return_tensors="pt").to(device)

    current_output = ""
    try:
        full_output_tokens = model.generate(
            **input_ids,
            max_new_tokens=150,
            temperature=0.7,
            do_sample=True,
            pad_token_id=tokenizer.eos_token_id
        )
        raw_output = tokenizer.decode(full_output_tokens[0][input_ids.input_ids.shape[1]:], skip_special_tokens=True)

        words = raw_output.split(" ")
        for i in range(len(words)):
            current_output += words[i] + " "
            yield gr.update(value=current_output) # Ensure outputs are gr.update
            time.sleep(STREAM_DELAY)
        return
    except Exception as e:
        yield gr.update(value=f"Error getting word meaning: {e}") # Ensure outputs are gr.update

# Function for Sentence Translation
def translate_sentence(sentence, target_language="Hindi"):
    if model is None or tokenizer is None:
        yield gr.update(value="Error: Model not loaded.")
        return
    if not sentence:
        yield gr.update(value="Please enter a sentence to translate.")
        return

    prompt = f"Translate the following English sentence into {target_language}:\n\nEnglish: {sentence}\n{target_language}:"
    messages = [{"role": "user", "content": prompt}]
    input_text_tokenized = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
    input_ids = tokenizer(input_text_tokenized, return_tensors="pt").to(device)

    current_output = ""
    try:
        full_output_tokens = model.generate(
            **input_ids,
            max_new_tokens=100,
            temperature=0.7,
            do_sample=True,
            pad_token_id=tokenizer.eos_token_id
        )
        raw_output = tokenizer.decode(full_output_tokens[0][input_ids.input_ids.shape[1]:], skip_special_tokens=True)

        words = raw_output.split(" ")
        for i in range(len(words)):
            current_output += words[i] + " "
            yield gr.update(value=current_output) # Ensure outputs are gr.update
            time.sleep(STREAM_DELAY)
        return
    except Exception as e:
        yield gr.update(value=f"Error translating sentence: {e}") # Ensure outputs are gr.update


# Initialize database
init_db()

# Define the chat functions OUTSIDE the Blocks for proper scoping
def user_message(user_message, history):
    # This now yields a gr.update to set the textbox value to empty
    # and updates the chatbot history to show the user's message immediately.
    return gr.update(value=""), history + [[user_message, None]]

def bot_response(history):
    if not history:
        # Yield an update for the chatbot to display an error
        yield gr.update(value=[["", "Error: Chat history is empty."]])
        return

    user_message_text = history[-1][0]

    # Generate the full response using the helper function
    full_bot_response = ""
    for chunk in generate_response(user_message_text, history[:-1]):
        full_bot_response = chunk
        history[-1][1] = full_bot_response
        yield history

def clear_chat():
    return None, None

# Make sure quiz_context = {} is defined globally at the top of your script, e.g.:
# quiz_context = {}

# --- Gradio UI ---
with gr.Blocks(title="Edu-Tutor AI", theme=gr.themes.Soft()) as demo:
    gr.Markdown("# <center>✨ Edu-Tutor AI: Your Personal Learning Assistant ✨</center>")
    gr.Markdown("Welcome! I'm Edu-Tutor, powered by IBM Granite. I can help you summarize text, refine your writing, understand new words, translate sentences, and even generate quizzes!")

    with gr.Tab("Chat with Edu-Tutor"):
        chatbot = gr.Chatbot(label="Edu-Tutor Chat", bubble_full_width=False, layout="panel", render_markdown=True, height=500)
        msg = gr.Textbox(label="Your message", placeholder="Ask me anything...", lines=2)
        with gr.Row():
            submit_button = gr.Button("Send Message")
            clear_button = gr.ClearButton([msg, chatbot])

        submit_button.click(user_message, [msg, chatbot], [msg, chatbot], queue=False).then(
            bot_response, chatbot, chatbot
        )
        msg.submit(user_message, [msg, chatbot], [msg, chatbot], queue=False).then(
            bot_response, chatbot, chatbot
        )
        clear_button.click(clear_chat, [], [msg, chatbot])

    with gr.Tab("Text Summarizer & Explainer"):
        with gr.Row():
            summary_input = gr.Textbox(label="Enter text to summarize/explain", lines=10, placeholder="Paste your text here...")
        with gr.Row():
            summary_type = gr.Radio(["short", "detailed", "simple_explanation"], label="Summary Type", value="short")
        with gr.Row():
            summarize_button = gr.Button("Generate Summary/Explanation")
        summary_output = gr.Markdown(label="Output")

        summarize_button.click(summarize_text, inputs=[summary_input, summary_type], outputs=summary_output)

    with gr.Tab("Text Refiner"):
        refine_input = gr.Textbox(label="Enter text to refine", lines=10, placeholder="Paste your text here...")
        refine_button = gr.Button("Refine Text")
        refine_output = gr.Markdown(label="Refined Text")

        refine_button.click(refine_text, inputs=refine_input, outputs=refine_output)

    with gr.Tab("Word Meaning & Usage"):
        word_input = gr.Textbox(label="Enter a word", placeholder="e.g., 'ubiquitous'")
        word_button = gr.Button("Get Meaning & Usage")
        word_output = gr.Markdown(label="Meaning and Examples")

        word_button.click(get_word_meaning_and_usage, inputs=word_input, outputs=word_output)

    with gr.Tab("Sentence Translator"):
        translate_input = gr.Textbox(label="Enter an English sentence", placeholder="e.g., 'Hello, how are you?'")
        target_language_dropdown = gr.Dropdown(["Hindi", "Spanish", "French", "German", "Japanese", "Tamil"], label="Translate to", value="Hindi")
        translate_button = gr.Button("Translate Sentence")
        translate_output = gr.Markdown(label="Translated Sentence")

        translate_button.click(translate_sentence, inputs=[translate_input, target_language_dropdown], outputs=translate_output)

    with gr.Tab("Generate Quiz"):
        gr.Markdown("## Generate a Multiple-Choice Quiz")
        gr.Markdown("Provide a topic or upload a PDF, and I'll generate a quiz for you.")
        with gr.Row():
            quiz_topic_input = gr.Textbox(label="Quiz Topic (e.g., 'Photosynthesis', 'World War II')", lines=1, placeholder="Enter a topic or upload a PDF")
            num_questions_input = gr.Slider(minimum=1, maximum=MAX_QUIZ_QUESTIONS_UI, value=3, step=1, label="Number of Questions")
            quiz_difficulty = gr.Radio(["easy", "medium", "hard"], label="Difficulty", value="medium")
        pdf_upload_input = gr.File(type="filepath", label="Upload PDF (Optional - overrides topic if provided)", file_types=[".pdf"])
        generate_quiz_button = gr.Button("Generate Quiz")

        # This Textbox is for debugging raw output. It will be hidden in production.
        raw_quiz_output_display = gr.Textbox(label="Raw Quiz Output (for debugging)", lines=10, visible=False)

        # State to store the parsed quiz data (This component is still needed, even if not directly an input to submit_quiz)
        parsed_quiz_data_state = gr.State(value=[])

        # Column to hold the interactive quiz questions
        with gr.Column(visible=False) as interactive_quiz_column:
            quiz_question_md = []
            quiz_options_radio = []
            for i in range(MAX_QUIZ_QUESTIONS_UI):
                quiz_question_md.append(gr.Markdown(value="", visible=False))
                quiz_options_radio.append(gr.Radio(choices=[], value=None, label="", visible=False))

            submit_quiz_button = gr.Button("Submit Quiz", visible=False)

        quiz_results_display = gr.Markdown(value="", visible=False, label="Quiz Results")
        start_new_quiz_button = gr.Button("Start a New Quiz", visible=False)

        # Define the outputs for generate_quiz_for_display
        # Order MUST match the `updates` list in the function.
        generate_quiz_outputs = [
            parsed_quiz_data_state, # State for quiz data
            raw_quiz_output_display, # Raw output for debugging
            submit_quiz_button, # Submit button visibility
            quiz_results_display, # Quiz results display (clear/hide)
            start_new_quiz_button, # Start new quiz button (hide)
            interactive_quiz_column # Interactive quiz column visibility
        ]
        # Dynamically add all question/radio components to outputs
        for i in range(MAX_QUIZ_QUESTIONS_UI):
            generate_quiz_outputs.append(quiz_question_md[i])
            generate_quiz_outputs.append(quiz_options_radio[i])

        generate_quiz_button.click(
            generate_quiz_for_display,
            inputs=[quiz_topic_input, num_questions_input, quiz_difficulty, pdf_upload_input],
            outputs=generate_quiz_outputs
        )

        # Define the outputs for submit_quiz
        # Order MUST match the `outputs` list in the function.
        submit_quiz_outputs = [
            submit_quiz_button,
            quiz_results_display,
            start_new_quiz_button,
            interactive_quiz_column # Hide the quiz column after submission
        ]
        # Dynamically add all question/radio components to outputs for hiding/clearing
        for i in range(MAX_QUIZ_QUESTIONS_UI):
            submit_quiz_outputs.append(quiz_question_md[i])
            submit_quiz_outputs.append(quiz_options_radio[i])

        submit_quiz_button.click(
            submit_quiz,
            inputs=list(quiz_options_radio), # Corrected: parsed_quiz_data_state removed
            outputs=submit_quiz_outputs
        )

        # Start New Quiz Button
        start_new_quiz_button.click(
            start_new_quiz_ui_reset,
            inputs=[],
            outputs=[quiz_topic_input, num_questions_input, quiz_difficulty, pdf_upload_input,
                     generate_quiz_button, raw_quiz_output_display, submit_quiz_button,
                     quiz_results_display, start_new_quiz_button, interactive_quiz_column] +
                     [item for sublist in zip(quiz_question_md, quiz_options_radio) for item in sublist]
        )

    with gr.Tab("Performance Dashboard"):
        gr.Markdown("## Your Learning Performance")
        gr.Markdown("Review your quiz history and identify weak areas.")

        with gr.Column():
            refresh_performance_button = gr.Button("Refresh Performance Data")
            weak_areas_display = gr.Markdown("### Weak Areas:\n\nNo data yet.")
            suggested_steps_display = gr.Markdown("### Suggested Next Steps:\n\nTake some quizzes to get personalized suggestions!")
            quiz_history_dataframe = gr.DataFrame(
                headers=["Timestamp", "Topic", "Difficulty", "Score", "Total Questions"],
                row_count=5, # Show at least 5 rows
                col_count=(5, "fixed"), # 5 fixed columns
                wrap=True,
                label="Quiz History",
                visible=True
            )

        # Load initial data when the tab is selected (or when refreshed)
        demo.load(
            get_performance_insights,
            inputs=[],
            outputs=[quiz_history_dataframe, weak_areas_display, suggested_steps_display]
        )

        # Refresh button click
        refresh_performance_button.click(
            get_performance_insights,
            inputs=[],
            outputs=[quiz_history_dataframe, weak_areas_display, suggested_steps_display]
        )

demo.launch(debug=True, share=True)