routes/questions.py

import os
from flask import current_app, request, jsonify, render_template, flash, redirect, url_for
from .common import main_bp, get_db_connection, login_required, current_user
from utils import sync_neetprep_collection
from processing import resize_image_if_needed, call_nim_ocr_api, extract_question_number_from_ocr_result
from strings import ROUTE_SAVE_QUESTIONS, ROUTE_EXTRACT_QUESTION_NUMBER, ROUTE_EXTRACT_ALL_QUESTION_NUMBERS, METHOD_POST
import requests
import json
from nvidia_prompts import BIOLOGY_PROMPT_TEMPLATE, CHEMISTRY_PROMPT_TEMPLATE, PHYSICS_PROMPT_TEMPLATE, MATHEMATICS_PROMPT_TEMPLATE, GENERAL_CLASSIFICATION_PROMPT
from utils import parse_note_payload

NVIDIA_API_KEY = os.getenv("NVIDIA_API_KEY")
NVIDIA_NIM_AVAILABLE = bool(NVIDIA_API_KEY)

def get_nvidia_prompt(subject, input_questions):
    if not subject or subject.lower() == 'auto':
        return GENERAL_CLASSIFICATION_PROMPT.format(input_questions=input_questions)
    if subject.lower() == 'biology': return BIOLOGY_PROMPT_TEMPLATE.format(input_questions=input_questions)
    if subject.lower() == 'chemistry': return CHEMISTRY_PROMPT_TEMPLATE.format(input_questions=input_questions)
    if subject.lower() == 'physics': return PHYSICS_PROMPT_TEMPLATE.format(input_questions=input_questions)
    if subject.lower() == 'mathematics': return MATHEMATICS_PROMPT_TEMPLATE.format(input_questions=input_questions)
    # Default to general prompt for unknown subjects
    return GENERAL_CLASSIFICATION_PROMPT.format(input_questions=input_questions)

@main_bp.route('/get_topic_suggestions', methods=['POST'])
@login_required
def get_topic_suggestions():
    data = request.json
    question_text, image_id, subject = data.get('question_text'), data.get('image_id'), data.get('subject')
    # Subject is now optional - will use general prompt if not provided
    if not question_text and image_id:
        try:
            conn = get_db_connection()
            # First check if we already have question_text cached
            q_row = conn.execute('SELECT question_text FROM questions WHERE image_id = ?', (image_id,)).fetchone()
            if q_row and q_row['question_text']:
                question_text = q_row['question_text']
            else:
                # Get image directly (works even if questions not saved yet)
                img_row = conn.execute('SELECT processed_filename, session_id FROM images WHERE id = ?', (image_id,)).fetchone()
                if img_row and img_row['processed_filename']:
                    image_path = os.path.join(current_app.config['PROCESSED_FOLDER'], img_row['processed_filename'])
                    if os.path.exists(image_path):
                        question_text = " ".join(item['text_prediction']['text'] for item in call_nim_ocr_api(resize_image_if_needed(image_path))['data'][0]['text_detections'])
                        # Cache the OCR result if question record exists
                        if q_row:
                            conn.execute('UPDATE questions SET question_text = ? WHERE image_id = ?', (question_text, image_id))
                            conn.commit()
            conn.close()
        except Exception as e: return jsonify({'error': f"OCR failed: {str(e)}"}), 500
    if not question_text: return jsonify({'error': 'Could not obtain question text.'}), 400

    prompt_content = get_nvidia_prompt(subject, f"1. {question_text}")
    if not NVIDIA_API_KEY: return jsonify({'error': 'NVIDIA_API_KEY not set'}), 500

    try:
        res = requests.post('https://integrate.api.nvidia.com/v1/chat/completions', headers={'Authorization': f'Bearer {NVIDIA_API_KEY}', 'Accept': 'application/json', 'Content-Type': 'application/json'}, json={"model": "nvidia/nemotron-3-nano-30b-a3b", "messages": [{"content": prompt_content, "role": "user"}], "temperature": 0.2, "top_p": 1, "max_tokens": 1024, "stream": False}, timeout=30)
        res.raise_for_status()
        content = res.json()['choices'][0]['message']['content']
        if "```json" in content: content = content.split("```json")[1].split("```")[0].strip()
        elif "```" in content: content = content.split("```")[1].split("```")[0].strip()
        parsed_data = json.loads(content)

        suggestions = []
        detected_subject = subject  # Default to input subject
        other_subjects = []

        if parsed_data.get('data'):
            item = parsed_data['data'][0]
            # Extract detected subject from AI response
            detected_subject = item.get('subject', subject) or subject
            # Extract other possible subjects
            other_subjects = item.get('other_possible_subjects', [])
            if isinstance(other_subjects, str):
                other_subjects = [other_subjects]

            primary = item.get('chapter_title')
            if primary and primary != 'Unclassified':
                suggestions.append(primary)
            if 'other_possible_chapters' in item:
                others = item['other_possible_chapters']
                if isinstance(others, list): suggestions.extend([c for c in others if c and c != 'Unclassified'])

        # Always return at least one suggestion with fallback
        if not suggestions:
            fallback_topics = {
                'biology': ['Cell: The Unit of Life', 'Biomolecules', 'Human Reproduction', 'Molecular Basis of Inheritance', 'Ecology'],
                'chemistry': ['Organic Chemistry – Some Basic Principles and Techniques (GOC)', 'Chemical Bonding and Molecular Structure', 'Thermodynamics', 'Electrochemistry'],
                'physics': ['Laws of Motion', 'Work, Energy and Power', 'Current Electricity', 'Electromagnetic Induction'],
                'mathematics': ['Calculus', 'Algebra', 'Coordinate Geometry', 'Probability']
            }
            subj_key = (detected_subject or 'biology').lower()
            suggestions = fallback_topics.get(subj_key, ['Unclassified'])

        return jsonify({
            'success': True,
            'suggestions': suggestions[:5],
            'subject': detected_subject,
            'other_possible_subjects': other_subjects,
            'full_response': parsed_data
        })
    except Exception as e:
        return jsonify({
            'success': True,
            'suggestions': ['Unclassified'],
            'subject': subject or 'Biology',
            'other_possible_subjects': [],
            'error': str(e)
        })

@main_bp.route('/get_topic_suggestions_batch', methods=['POST'])
@login_required
def get_topic_suggestions_batch():
    """Batch endpoint for getting topic suggestions for multiple images at once.
    Requires a subject to be specified. Processes up to 8 questions in a single API call."""
    data = request.json
    image_ids = data.get('image_ids', [])
    subject = data.get('subject')

    current_app.logger.info(f"[BATCH] Received request for {len(image_ids)} images, subject={subject}")

    if not image_ids:
        return jsonify({'error': 'No image_ids provided'}), 400
    if not subject or subject.lower() == 'auto':
        return jsonify({'error': 'Subject must be specified for batch requests'}), 400
    if len(image_ids) > 8:
        return jsonify({'error': 'Maximum 8 images per batch'}), 400

    if not NVIDIA_API_KEY:
        return jsonify({'error': 'NVIDIA_API_KEY not set'}), 500

    try:
        conn = get_db_connection()
        questions_data = []  # List of {image_id, index, question_text}

        for idx, image_id in enumerate(image_ids):
            # Verify ownership
            owner = conn.execute("SELECT s.user_id FROM images i JOIN sessions s ON i.session_id = s.id WHERE i.id = ?", (image_id,)).fetchone()
            if not owner or owner['user_id'] != current_user.id:
                current_app.logger.warning(f"[BATCH] Skipping image {image_id}: unauthorized")
                continue  # Skip unauthorized images

            question_text = None
            # First check if we already have question_text cached
            q_row = conn.execute('SELECT question_text FROM questions WHERE image_id = ?', (image_id,)).fetchone()
            if q_row and q_row['question_text']:
                question_text = q_row['question_text']
                current_app.logger.debug(f"[BATCH] Image {image_id}: using cached question_text")
            else:
                # Get image directly
                img_row = conn.execute('SELECT processed_filename FROM images WHERE id = ?', (image_id,)).fetchone()
                if img_row and img_row['processed_filename']:
                    image_path = os.path.join(current_app.config['PROCESSED_FOLDER'], img_row['processed_filename'])
                    if os.path.exists(image_path):
                        try:
                            current_app.logger.info(f"[BATCH] Image {image_id}: running OCR on {img_row['processed_filename']}")
                            question_text = " ".join(item['text_prediction']['text'] for item in call_nim_ocr_api(resize_image_if_needed(image_path))['data'][0]['text_detections'])
                            # Cache the OCR result if question record exists
                            if q_row:
                                conn.execute('UPDATE questions SET question_text = ? WHERE image_id = ?', (question_text, image_id))
                                conn.commit()
                        except Exception as ocr_err:
                            current_app.logger.error(f"[BATCH] Image {image_id}: OCR failed - {ocr_err}")
                    else:
                        current_app.logger.warning(f"[BATCH] Image {image_id}: file not found at {image_path}")
                else:
                    current_app.logger.warning(f"[BATCH] Image {image_id}: no processed_filename in DB")

            if question_text:
                questions_data.append({'image_id': image_id, 'index': idx + 1, 'question_text': question_text})
            else:
                current_app.logger.warning(f"[BATCH] Image {image_id}: no question_text obtained")

        conn.close()

        current_app.logger.info(f"[BATCH] Got question text for {len(questions_data)}/{len(image_ids)} images")

        if not questions_data:
            return jsonify({'error': 'Could not obtain question text for any images'}), 400

        # Build multi-question prompt
        input_questions = "\n".join(f"{q['index']}. {q['question_text']}" for q in questions_data)
        prompt_content = get_nvidia_prompt(subject, input_questions)

        current_app.logger.info(f"[BATCH] Sending {len(questions_data)} questions to NVIDIA API")
        current_app.logger.debug(f"[BATCH] Prompt preview: {input_questions[:500]}...")

        # Make single API call for all questions
        res = requests.post(
            'https://integrate.api.nvidia.com/v1/chat/completions',
            headers={'Authorization': f'Bearer {NVIDIA_API_KEY}', 'Accept': 'application/json', 'Content-Type': 'application/json'},
            json={"model": "nvidia/nemotron-3-nano-30b-a3b", "messages": [{"content": prompt_content, "role": "user"}], "temperature": 0.2, "top_p": 1, "max_tokens": 2048, "stream": False},
            timeout=60
        )
        res.raise_for_status()
        content = res.json()['choices'][0]['message']['content']

        current_app.logger.info(f"[BATCH] NVIDIA API response length: {len(content)} chars")
        current_app.logger.debug(f"[BATCH] Raw response: {content[:1000]}...")

        # Parse JSON from response
        if "```json" in content:
            content = content.split("```json")[1].split("```")[0].strip()
        elif "```" in content:
            content = content.split("```")[1].split("```")[0].strip()
        parsed_data = json.loads(content)

        current_app.logger.info(f"[BATCH] Parsed data has {len(parsed_data.get('data', []))} items")

        # Build results for each image
        results = {}
        fallback_topics = {
            'biology': ['Cell: The Unit of Life', 'Biomolecules', 'Human Reproduction'],
            'chemistry': ['Organic Chemistry – Some Basic Principles and Techniques (GOC)', 'Chemical Bonding and Molecular Structure'],
            'physics': ['Laws of Motion', 'Work, Energy and Power', 'Current Electricity'],
            'mathematics': ['Calculus', 'Algebra', 'Coordinate Geometry']
        }

        if parsed_data.get('data'):
            data_items = parsed_data['data']
            current_app.logger.info(f"[BATCH] AI returned {len(data_items)} items, we have {len(questions_data)} questions")

            # Try to match by index first, then fall back to order-based matching
            matched_by_index = 0
            for item in data_items:
                item_index = item.get('index', 0)
                current_app.logger.debug(f"[BATCH] Processing item with index={item_index}: {item}")
                # Find the matching question by our sequential index
                matching_q = next((q for q in questions_data if q['index'] == item_index), None)
                if matching_q:
                    matched_by_index += 1
                    suggestions = []
                    primary = item.get('chapter_title')
                    if primary and primary != 'Unclassified':
                        suggestions.append(primary)
                    if 'other_possible_chapters' in item:
                        others = item['other_possible_chapters']
                        if isinstance(others, list):
                            suggestions.extend([c for c in others if c and c != 'Unclassified'])

                    if not suggestions:
                        suggestions = fallback_topics.get(subject.lower(), ['Unclassified'])

                    results[matching_q['image_id']] = {
                        'success': True,
                        'suggestions': suggestions[:5],
                        'subject': subject,
                        'other_possible_subjects': []
                    }
                    current_app.logger.info(f"[BATCH] Image {matching_q['image_id']}: matched by index, suggestions={suggestions[:3]}")

            # If index matching failed, try order-based matching
            if matched_by_index == 0 and len(data_items) > 0:
                current_app.logger.warning(f"[BATCH] Index matching failed, trying order-based matching")
                for i, item in enumerate(data_items):
                    if i < len(questions_data):
                        q = questions_data[i]
                        if q['image_id'] not in results:
                            suggestions = []
                            primary = item.get('chapter_title')
                            if primary and primary != 'Unclassified':
                                suggestions.append(primary)
                            if 'other_possible_chapters' in item:
                                others = item['other_possible_chapters']
                                if isinstance(others, list):
                                    suggestions.extend([c for c in others if c and c != 'Unclassified'])

                            if not suggestions:
                                suggestions = fallback_topics.get(subject.lower(), ['Unclassified'])

                            results[q['image_id']] = {
                                'success': True,
                                'suggestions': suggestions[:5],
                                'subject': subject,
                                'other_possible_subjects': []
                            }
                            current_app.logger.info(f"[BATCH] Image {q['image_id']}: matched by order, suggestions={suggestions[:3]}")

        # Fill in any missing results
        for q in questions_data:
            if q['image_id'] not in results:
                current_app.logger.warning(f"[BATCH] Image {q['image_id']}: using fallback (no match in API response)")
                results[q['image_id']] = {
                    'success': True,
                    'suggestions': fallback_topics.get(subject.lower(), ['Unclassified']),
                    'subject': subject,
                    'other_possible_subjects': []
                }

        return jsonify({'success': True, 'results': results})

    except Exception as e:
        # Return error but with fallback results for each image
        fallback_result = {
            'success': True,
            'suggestions': ['Unclassified'],
            'subject': subject,
            'other_possible_subjects': [],
            'error': str(e)
        }
        return jsonify({
            'success': False,
            'error': str(e),
            'results': {img_id: fallback_result for img_id in image_ids}
        })

@main_bp.route('/classified/update_single', methods=['POST'])
@login_required
def update_question_classification_single():
    data = request.json
    image_id, subject, chapter = data.get('image_id'), data.get('subject'), data.get('chapter')
    if not image_id: return jsonify({'error': 'Image ID is required'}), 400
    try:
        conn = get_db_connection()
        # Get image info and verify ownership
        img_info = conn.execute("""
            SELECT i.session_id, s.user_id FROM images i
            JOIN sessions s ON i.session_id = s.id WHERE i.id = ?
        """, (image_id,)).fetchone()
        if not img_info or img_info['user_id'] != current_user.id:
            conn.close(); return jsonify({'error': 'Unauthorized'}), 403

        # Check if question row exists
        existing = conn.execute('SELECT id FROM questions WHERE image_id = ?', (image_id,)).fetchone()
        if existing:
            conn.execute('UPDATE questions SET subject = ?, chapter = ? WHERE image_id = ?', (subject, chapter, image_id))
        else:
            # Create question row if it doesn't exist
            conn.execute('''
                INSERT INTO questions (session_id, image_id, question_number, subject, chapter, status)
                VALUES (?, ?, '', ?, ?, 'unattempted')
            ''', (img_info['session_id'], image_id, subject, chapter))

        # Auto-convert and sync neetprep collection
        sync_neetprep_collection(conn, session_id, current_user.id)

        conn.commit(); conn.close()
        return jsonify({'success': True})
    except Exception as e:
        current_app.logger.error(f"Error updating classification: {e}")
        return jsonify({'error': str(e)}), 500


@main_bp.route('/question_entry_v2/<session_id>')
@login_required
def question_entry_v2(session_id):
    conn = get_db_connection()
    session_data = conn.execute('SELECT original_filename, subject, tags, notes FROM sessions WHERE id = ? AND user_id = ?', (session_id, current_user.id)).fetchone()
    if not session_data:
        conn.close(); flash("Session not found or you don't have permission to access it.", "warning"); return redirect(url_for('dashboard.dashboard'))
    images = conn.execute("""SELECT i.id, i.processed_filename, i.note_filename, i.note_json, i.include_note_in_pdf, q.question_number, q.status, q.marked_solution, q.actual_solution
           FROM images i
           LEFT JOIN questions q ON i.id = q.image_id
           WHERE i.session_id = ? AND i.image_type = 'cropped'
           ORDER BY i.id""", (session_id,)).fetchall()
    classified_count = conn.execute("""SELECT COUNT(*) as count FROM images i LEFT JOIN questions q ON i.id = q.image_id WHERE i.session_id = ? AND i.image_type = 'cropped' AND q.subject IS NOT NULL AND q.chapter IS NOT NULL""", (session_id,)).fetchone()['count']
    conn.close()
    if not images: return "No questions were created from the PDF. Please go back and draw crop boxes.", 404

    image_dicts = []
    for img in images:
        img_dict = dict(img)
        note_payload = parse_note_payload(img_dict.get('note_json'))
        linked_pages = note_payload.get('linked_pdf_pages', [])
        img_dict['linked_pdf_pages'] = linked_pages
        img_dict['linked_pdf_count'] = len(linked_pages)
        img_dict['has_note'] = bool(img_dict.get('note_filename') or img_dict.get('note_json') or linked_pages)
        image_dicts.append(img_dict)

    return render_template('question_entry_v2.html', session_id=session_id, images=image_dicts, session_data=dict(session_data) if session_data else {}, classified_count=classified_count, total_questions=len(images), nvidia_nim_available=NVIDIA_NIM_AVAILABLE)

@main_bp.route(ROUTE_SAVE_QUESTIONS, methods=[METHOD_POST])
@login_required
def save_questions():
    data = request.json
    session_id, questions = data['session_id'], data['questions']
    conn = get_db_connection()
    session_owner = conn.execute('SELECT user_id FROM sessions WHERE id = ?', (session_id,)).fetchone()
    if not session_owner or session_owner['user_id'] != current_user.id:
        conn.close(); return jsonify({'error': 'Unauthorized'}), 403
    conn.execute('UPDATE sessions SET subject = ?, tags = ?, notes = ? WHERE id = ?', (data.get('pdf_subject', ''), data.get('pdf_tags', ''), data.get('pdf_notes', ''), session_id))

    # Use UPSERT to preserve subject and chapter data from classification
    for q in questions:
        existing = conn.execute('SELECT id, subject, chapter FROM questions WHERE image_id = ?', (q['image_id'],)).fetchone()
        if existing:
            # Update existing, preserve subject and chapter
            conn.execute('''
                UPDATE questions
                SET question_number = ?, status = ?, marked_solution = ?, actual_solution = ?, tags = ?
                WHERE image_id = ?
            ''', (q['question_number'], q['status'], q.get('marked_solution', ''), q.get('actual_solution', ''), data.get('pdf_tags', ''), q['image_id']))
        else:
            # Insert new
            conn.execute('''
                INSERT INTO questions (session_id, image_id, question_number, subject, status, marked_solution, actual_solution, time_taken, tags)
                VALUES (?, ?, ?, '', ?, ?, ?, ?, ?)
            ''', (session_id, q['image_id'], q['question_number'], q['status'], q.get('marked_solution', ''), q.get('actual_solution', ''), q.get('time_taken', ''), data.get('pdf_tags', '')))

    # Auto-convert and sync neetprep collection
    sync_neetprep_collection(conn, session_id, current_user.id)

    conn.commit(); conn.close()
    return jsonify({'success': True, 'message': 'Questions saved successfully.'})

@main_bp.route('/autosave_question', methods=['POST'])
@login_required
def autosave_question():
    """Auto-save a single question's data without affecting other questions."""
    data = request.json
    session_id = data.get('session_id')
    question = data.get('question', {})

    if not session_id or not question.get('image_id'):
        return jsonify({'error': 'Missing session_id or image_id'}), 400

    conn = get_db_connection()
    try:
        # Verify session ownership
        session_owner = conn.execute('SELECT user_id FROM sessions WHERE id = ?', (session_id,)).fetchone()
        if not session_owner or session_owner['user_id'] != current_user.id:
            conn.close()
            return jsonify({'error': 'Unauthorized'}), 403

        image_id = question['image_id']
        question_number = question.get('question_number', '')
        status = question.get('status', 'unattempted')
        marked_solution = question.get('marked_solution', '')
        actual_solution = question.get('actual_solution', '')

        # Check if question already exists
        existing = conn.execute('SELECT id, subject, chapter, tags FROM questions WHERE image_id = ?', (image_id,)).fetchone()

        if existing:
            # Update existing question, preserving subject and chapter
            conn.execute('''
                UPDATE questions
                SET question_number = ?, status = ?, marked_solution = ?, actual_solution = ?
                WHERE image_id = ?
            ''', (question_number, status, marked_solution, actual_solution, image_id))
        else:
            # Insert new question
            conn.execute('''
                INSERT INTO questions (session_id, image_id, question_number, subject, status, marked_solution, actual_solution, time_taken, tags)
                VALUES (?, ?, ?, '', ?, ?, ?, '', '')
            ''', (session_id, image_id, question_number, status, marked_solution, actual_solution))

        # Auto-convert and sync neetprep collection
        sync_neetprep_collection(conn, session_id, current_user.id)

        conn.commit()
        conn.close()
        return jsonify({'success': True})

    except Exception as e:
        conn.close()
        current_app.logger.error(f"Auto-save question error: {e}")
        return jsonify({'error': str(e)}), 500

@main_bp.route('/autosave_session_metadata', methods=['POST'])
@login_required
def autosave_session_metadata():
    """Auto-save session metadata (PDF subject, tags, notes)."""
    data = request.json
    session_id = data.get('session_id')

    if not session_id:
        return jsonify({'error': 'Missing session_id'}), 400

    conn = get_db_connection()
    try:
        # Verify session ownership
        session_owner = conn.execute('SELECT user_id FROM sessions WHERE id = ?', (session_id,)).fetchone()
        if not session_owner or session_owner['user_id'] != current_user.id:
            conn.close()
            return jsonify({'error': 'Unauthorized'}), 403

        # Update session metadata
        conn.execute('''
            UPDATE sessions
            SET subject = ?, tags = ?, notes = ?
            WHERE id = ?
        ''', (
            data.get('pdf_subject', ''),
            data.get('pdf_tags', ''),
            data.get('pdf_notes', ''),
            session_id
        ))

        conn.commit()
        conn.close()
        return jsonify({'success': True})

    except Exception as e:
        conn.close()
        current_app.logger.error(f"Auto-save metadata error: {e}")
        return jsonify({'error': str(e)}), 500

@main_bp.route(ROUTE_EXTRACT_QUESTION_NUMBER, methods=[METHOD_POST])
@login_required
def extract_question_number():
    if not NVIDIA_NIM_AVAILABLE: return jsonify({'error': 'NVIDIA NIM feature is not available.'}), 400
    data = request.json
    image_id = data.get('image_id')
    if not image_id: return jsonify({'error': 'Missing image_id parameter'}), 400
    try:
        conn = get_db_connection()
        image_owner = conn.execute("SELECT s.user_id FROM images i JOIN sessions s ON i.session_id = s.id WHERE i.id = ?", (image_id,)).fetchone()
        if not image_owner or image_owner['user_id'] != current_user.id:
            conn.close(); return jsonify({'error': 'Unauthorized'}), 403
        image_info = conn.execute('SELECT processed_filename FROM images WHERE id = ?', (image_id,)).fetchone()
        conn.close()
        if not image_info or not image_info['processed_filename']: return jsonify({'error': 'Image not found or not processed'}), 404
        image_path = os.path.join(current_app.config['PROCESSED_FOLDER'], image_info['processed_filename'])
        if not os.path.exists(image_path): return jsonify({'error': 'Image file not found on disk'}), 404
        image_bytes = resize_image_if_needed(image_path)
        ocr_result = call_nim_ocr_api(image_bytes)
        question_number = extract_question_number_from_ocr_result(ocr_result)
        return jsonify({'success': True, 'question_number': question_number, 'image_id': image_id})
    except Exception as e: return jsonify({'error': f'Failed to extract question number: {str(e)}'}), 500

@main_bp.route(ROUTE_EXTRACT_ALL_QUESTION_NUMBERS, methods=[METHOD_POST])
@login_required
def extract_all_question_numbers():
    if not NVIDIA_NIM_AVAILABLE: return jsonify({'error': 'NVIDIA NIM feature is not available.'}), 400
    data = request.json
    session_id = data.get('session_id')
    if not session_id: return jsonify({'error': 'Missing session_id parameter'}), 400
    try:
        conn = get_db_connection()
        session_owner = conn.execute('SELECT user_id FROM sessions WHERE id = ?', (session_id,)).fetchone()
        if not session_owner or session_owner['user_id'] != current_user.id:
            conn.close(); return jsonify({'error': 'Unauthorized'}), 403
        images = conn.execute("SELECT id, processed_filename FROM images WHERE session_id = ? AND image_type = 'cropped' ORDER BY id", (session_id,)).fetchall()
        conn.close()
        if not images: return jsonify({'error': 'No cropped images found in session'}), 404
        results, errors = [], []
        MAX_CONCURRENT_REQUESTS = 5
        processed_count = 0
        for image in images:
            if processed_count >= MAX_CONCURRENT_REQUESTS:
                import time; time.sleep(1); processed_count = 0
            try:
                image_id, processed_filename = image['id'], image['processed_filename']
                if not processed_filename:
                    errors.append({'image_id': image_id, 'error': 'Image not processed'}); continue
                image_path = os.path.join(current_app.config['PROCESSED_FOLDER'], processed_filename)
                if not os.path.exists(image_path):
                    errors.append({'image_id': image_id, 'error': 'Image file not found on disk'}); continue
                image_bytes = resize_image_if_needed(image_path)
                ocr_result = call_nim_ocr_api(image_bytes)
                question_number = extract_question_number_from_ocr_result(ocr_result)
                results.append({'image_id': image_id, 'question_number': question_number})
                processed_count += 1
            except Exception as e: errors.append({'image_id': image['id'], 'error': str(e)})
        return jsonify({'success': True, 'results': results, 'errors': errors})
    except Exception as e: return jsonify({'error': f'Failed to extract question numbers: {str(e)}'}), 500

@main_bp.route('/delete_question/<image_id>', methods=['DELETE'])
@login_required
def delete_question(image_id):
    try:
        conn = get_db_connection()
        image_owner = conn.execute("SELECT s.user_id FROM images i JOIN sessions s ON i.session_id = s.id WHERE i.id = ?", (image_id,)).fetchone()
        if not image_owner or image_owner['user_id'] != current_user.id:
            conn.close(); return jsonify({'error': 'Unauthorized'}), 403
        image_info = conn.execute('SELECT session_id, filename, processed_filename FROM images WHERE id = ?', (image_id,)).fetchone()
        if not image_info: conn.close(); return jsonify({'error': 'Question not found'}), 404
        # Get question ID to delete bookmarks
        q_row = conn.execute('SELECT id FROM questions WHERE image_id = ?', (image_id,)).fetchone()
        if q_row:
            conn.execute('DELETE FROM neetprep_bookmarks WHERE neetprep_question_id = ? AND question_type = ?', (str(q_row['id']), 'classified'))
        
        conn.execute('DELETE FROM questions WHERE image_id = ?', (image_id,))
        conn.execute('DELETE FROM images WHERE id = ?', (image_id,))
        
        # Auto-convert and sync neetprep collection
        sync_neetprep_collection(conn, image_info['session_id'], current_user.id)

        conn.commit(); conn.close()
        return jsonify({'success': True})
    except Exception as e: return jsonify({'error': str(e)}), 500