app.py

import os
import tempfile

# Fix cache permissions for Hugging Face Spaces - MUST be at the very top
os.environ['TRANSFORMERS_CACHE'] = '/tmp/transformers_cache'
os.environ['HF_HOME'] = '/tmp/huggingface_cache'
os.environ['TORCH_HOME'] = '/tmp/torch_cache'
os.environ['SENTENCE_TRANSFORMERS_HOME'] = '/tmp/sentence_transformers'

# Create cache directories
cache_dirs = ['/tmp/transformers_cache', '/tmp/huggingface_cache', '/tmp/torch_cache', '/tmp/sentence_transformers']
for cache_dir in cache_dirs:
    os.makedirs(cache_dir, exist_ok=True)

from flask import Flask, render_template, request, jsonify
import json
from sentence_transformers import SentenceTransformer
import numpy as np
from sklearn.metrics.pairwise import cosine_similarity
import re
from transformers import pipeline
import torch
import nltk
from collections import Counter
import unicodedata

# Download required NLTK data
try:
    nltk.download('punkt', quiet=True)
    nltk.download('wordnet', quiet=True)
    nltk.download('stopwords', quiet=True)
except:
    print("Warning: Could not download NLTK data")

app = Flask(__name__)


class EvaluationMetrics:
    """Class to calculate F1, BLEU, and ROUGE-L scores"""
    
    @staticmethod
    def normalize_text(text):
        """Normalize text for comparison"""
        # Remove extra whitespace and normalize unicode
        text = unicodedata.normalize('NFKD', text)
        text = ' '.join(text.split())
        return text.lower()
    
    @staticmethod
    def tokenize_text(text):
        """Simple tokenization for Hindi/English mixed text"""
        # Simple tokenization that works for both Hindi and English
        text = re.sub(r'[^\w\s]', ' ', text)
        return text.split()
    
    @staticmethod
    def calculate_f1_score(predicted, reference):
        """Calculate F1 score between predicted and reference text"""
        try:
            pred_tokens = set(EvaluationMetrics.tokenize_text(EvaluationMetrics.normalize_text(predicted)))
            ref_tokens = set(EvaluationMetrics.tokenize_text(EvaluationMetrics.normalize_text(reference)))
            
            if len(pred_tokens) == 0 and len(ref_tokens) == 0:
                return 1.0
            
            if len(pred_tokens) == 0 or len(ref_tokens) == 0:
                return 0.0
            
            # Calculate intersection
            common_tokens = pred_tokens.intersection(ref_tokens)
            
            if len(common_tokens) == 0:
                return 0.0
            
            precision = len(common_tokens) / len(pred_tokens)
            recall = len(common_tokens) / len(ref_tokens)
            
            f1 = 2 * (precision * recall) / (precision + recall)
            return float(f1)  # Convert to Python float
        except:
            return 0.0
    
    @staticmethod
    def calculate_bleu_score(predicted, reference):
        """Calculate BLEU score"""
        try:
            from nltk.translate.bleu_score import sentence_bleu, SmoothingFunction
            
            pred_tokens = EvaluationMetrics.tokenize_text(EvaluationMetrics.normalize_text(predicted))
            ref_tokens = EvaluationMetrics.tokenize_text(EvaluationMetrics.normalize_text(reference))
            
            if len(pred_tokens) == 0 or len(ref_tokens) == 0:
                return 0.0
            
            # Use smoothing function to handle zero n-grams
            smoothing = SmoothingFunction()
            bleu_score = sentence_bleu(
                [ref_tokens], 
                pred_tokens, 
                smoothing_function=smoothing.method1
            )
            
            return float(bleu_score)  # Convert to Python float
        except:
            # Fallback BLEU calculation if NLTK fails
            return EvaluationMetrics.simple_bleu(predicted, reference)
    
    @staticmethod
    def simple_bleu(predicted, reference):
        """Simple BLEU calculation fallback"""
        try:
            pred_tokens = EvaluationMetrics.tokenize_text(EvaluationMetrics.normalize_text(predicted))
            ref_tokens = EvaluationMetrics.tokenize_text(EvaluationMetrics.normalize_text(reference))
            
            if len(pred_tokens) == 0 or len(ref_tokens) == 0:
                return 0.0
            
            # Calculate 1-gram precision (simplified BLEU)
            pred_counts = Counter(pred_tokens)
            ref_counts = Counter(ref_tokens)
            
            overlap = sum(min(pred_counts[token], ref_counts[token]) for token in pred_counts)
            precision = overlap / len(pred_tokens) if len(pred_tokens) > 0 else 0
            
            return float(precision)  # Convert to Python float
        except:
            return 0.0
    
    @staticmethod
    def lcs_length(x, y):
        """Calculate Longest Common Subsequence length"""
        m, n = len(x), len(y)
        dp = [[0] * (n + 1) for _ in range(m + 1)]
        
        for i in range(1, m + 1):
            for j in range(1, n + 1):
                if x[i-1] == y[j-1]:
                    dp[i][j] = dp[i-1][j-1] + 1
                else:
                    dp[i][j] = max(dp[i-1][j], dp[i][j-1])
        
        return dp[m][n]
    
    @staticmethod
    def calculate_rouge_l(predicted, reference):
        """Calculate ROUGE-L score"""
        try:
            pred_tokens = EvaluationMetrics.tokenize_text(EvaluationMetrics.normalize_text(predicted))
            ref_tokens = EvaluationMetrics.tokenize_text(EvaluationMetrics.normalize_text(reference))
            
            if len(pred_tokens) == 0 and len(ref_tokens) == 0:
                return 1.0
            
            if len(pred_tokens) == 0 or len(ref_tokens) == 0:
                return 0.0
            
            lcs_len = EvaluationMetrics.lcs_length(pred_tokens, ref_tokens)
            
            if lcs_len == 0:
                return 0.0
            
            precision = lcs_len / len(pred_tokens)
            recall = lcs_len / len(ref_tokens)
            
            if precision + recall == 0:
                return 0.0
            
            rouge_l = 2 * precision * recall / (precision + recall)
            return float(rouge_l)  # Convert to Python float
        except:
            return 0.0
    
    @staticmethod
    def calculate_all_scores(predicted, reference):
        """Calculate all evaluation metrics"""
        if not predicted or not reference:
            return {
                'f1_score': 0.0,
                'bleu_score': 0.0,
                'rouge_l_score': 0.0
            }
        
        return {
            'f1_score': round(EvaluationMetrics.calculate_f1_score(predicted, reference), 4),
            'bleu_score': round(EvaluationMetrics.calculate_bleu_score(predicted, reference), 4),
            'rouge_l_score': round(EvaluationMetrics.calculate_rouge_l(predicted, reference), 4)
        }


class ImprovedVATIKAChatbot:
    def __init__(self):
        print("🚀 Initializing KashiVani Chatbot...")

        # Load multilingual embedding model
        self.embedding_model = SentenceTransformer('sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2')
        print("✅ Embedding model loaded")

        # Load QA model with better error handling
        self.qa_pipeline = self.load_qa_model()

        # Initialize data structures
        self.contexts = []
        self.context_embeddings = None
        self.qa_pairs = []  # Store all Q&A pairs separately
        self.qa_embeddings = None

        # Initialize evaluation metrics
        self.evaluation_metrics = EvaluationMetrics()

        # Load and process data
        self.load_data()
        print(f"✅ Loaded {len(self.contexts)} contexts and {len(self.qa_pairs)} Q&A pairs")

    def load_qa_model(self):
        """Load QA model with fallback options"""
        models_to_try = [
            "deepset/xlm-roberta-base-squad2",
            "distilbert-base-multilingual-cased",
            "deepset/minilm-uncased-squad2"
        ]

        for model_name in models_to_try:
            try:
                print(f"🔄 Trying to load QA model: {model_name}")
                qa_pipeline = pipeline(
                    "question-answering",
                    model=model_name,
                    tokenizer=model_name,
                    device=-1  # Use CPU
                )
                print(f"✅ Successfully loaded: {model_name}")
                return qa_pipeline
            except Exception as e:
                print(f"❌ Failed to load {model_name}: {e}")
                continue

        print("⚠️ Could not load any QA model, using fallback")
        return None

    def load_data(self):
        """Load and preprocess VATIKA dataset with better error handling"""
        try:
            # Check if data files exist
            train_file = 'data/train.json'
            val_file = 'data/validation.json'

            if not os.path.exists(train_file):
                print("❌ Train file not found, creating sample data...")
                self.create_sample_data()

            # Load training data
            with open(train_file, 'r', encoding='utf-8') as f:
                train_data = json.load(f)

            all_data = train_data.get('domains', [])

            # Try to load validation data
            if os.path.exists(val_file):
                try:
                    with open(val_file, 'r', encoding='utf-8') as f:
                        val_data = json.load(f)
                    all_data.extend(val_data.get('domains', []))
                    print("✅ Validation data loaded")
                except Exception as e:
                    print(f"⚠️-- Could not load validation data: {e}")

            # Process data
            self.process_data(all_data)

        except Exception as e:
            print(f"❌ Error loading data: {e}")
            self.create_fallback_data()

    def process_data(self, domains_data):
        """Process loaded data and create embeddings"""
        all_contexts = []
        all_qas = []

        for domain_data in domains_data:
            domain = domain_data.get('domain', 'unknown')

            for context_data in domain_data.get('contexts', []):
                context_text = context_data.get('context', '')
                qas = context_data.get('qas', [])

                if context_text.strip():  # Only add non-empty contexts
                    context_info = {
                        'domain': domain,
                        'context': context_text,
                        'qas': qas
                    }
                    all_contexts.append(context_info)

                    # Extract Q&A pairs
                    for qa in qas:
                        question = qa.get('question', '').strip()
                        answer = qa.get('answer', '').strip()

                        if question and answer:
                            qa_info = {
                                'question': question,
                                'answer': answer,
                                'domain': domain,
                                'context': context_text
                            }
                            all_qas.append(qa_info)

        self.contexts = all_contexts
        self.qa_pairs = all_qas

        # Create embeddings
        if self.contexts:
            print("🔄 Creating context embeddings...")
            context_texts = [ctx['context'] for ctx in self.contexts]
            self.context_embeddings = self.embedding_model.encode(context_texts, show_progress_bar=True)

        if self.qa_pairs:
            print("🔄 Creating Q&A embeddings...")
            qa_questions = [qa['question'] for qa in self.qa_pairs]
            self.qa_embeddings = self.embedding_model.encode(qa_questions, show_progress_bar=True)

    def create_sample_data(self):
        """Create sample data if original data is not available"""
        sample_data = {
            "domains": [
                {
                    "domain": "varanasi_temples",
                    "contexts": [
                        {
                            "context": "काशी विश्वनाथ मंदिर वाराणसी का सबसे प्रसिद्ध और पवित्र मंदिर है। यह भगवान शिव को समर्पित है और गंगा नदी के पश्चिमी तट पर स्थित है। यह 12 ज्योतिर्लिंगों में से एक है और हिंदू धर्म में अत्यंत महत्वपूर्ण माना जाता है।",
                            "qas": [
                                {
                                    "question": "काशी विश्वनाथ मंदिर कहाँ स्थित है?",
                                    "answer": "काशी विश्वनाथ मंदिर वाराणसी में गंगा नदी के पश्चिमी तट पर स्थित है।"
                                },
                                {
                                    "question": "काशी विश्वनाथ मंदिर किसे समर्पित है?",
                                    "answer": "काशी विश्वनाथ मंदिर भगवान शिव को समर्पित है।"
                                },
                                {
                                    "question": "क्या काशी विश्वनाथ ज्योतिर्लिंग है?",
                                    "answer": "हाँ, काशी विश्वनाथ मंदिर 12 ज्योतिर्लिंगों में से एक है।"
                                }
                            ]
                        }
                    ]
                },
                {
                    "domain": "varanasi_ghats",
                    "contexts": [
                        {
                            "context": "दशाश्वमेध घाट वाराणसी का सबसे प्रसिद्ध और मुख्य घाट है। यहाँ प्रतिदिन शाम को भव्य गंगा आरती का आयोजन होता है। यह घाट अत्यंत पवित्र माना जाता है और हजारों श्रद्धालु और पर्यटक यहाँ आते हैं।",
                            "qas": [
                                {
                                    "question": "दशाश्वमेध घाट पर आरती कब होती है?",
                                    "answer": "दशाश्वमेध घाट पर प्रतिदिन शाम को गंगा आरती होती है।"
                                },
                                {
                                    "question": "वाराणसी का सबसे प्रसिद्ध घाट कौन सा है?",
                                    "answer": "दशाश्वमेध घाट वाराणसी का सबसे प्रसिद्ध घाट है।"
                                }
                            ]
                        }
                    ]
                }
            ]
        }

        os.makedirs('data', exist_ok=True)
        with open('data/train.json', 'w', encoding='utf-8') as f:
            json.dump(sample_data, f, ensure_ascii=False, indent=2)

        print("✅ Sample data created")

    def create_fallback_data(self):
        """Create minimal fallback data"""
        self.contexts = [{
            'domain': 'general',
            'context': 'वाराणसी भारत का एक प्राचीन और पवित्र शहर है।',
            'qas': [{'question': 'वाराणसी क्या है?', 'answer': 'वाराणसी भारत का एक प्राचीन और पवित्र शहर है।'}]
        }]

        self.qa_pairs = [{'question': 'वाराणसी क्या है?', 'answer': 'वाराणसी भारत का एक प्राचीन और पवित्र शहर है।',
                          'domain': 'general'}]

        context_texts = [ctx['context'] for ctx in self.contexts]
        self.context_embeddings = self.embedding_model.encode(context_texts)

        qa_questions = [qa['question'] for qa in self.qa_pairs]
        self.qa_embeddings = self.embedding_model.encode(qa_questions)

    def find_best_qa_match(self, query, threshold=0.6):
        """Find best matching Q&A pair"""
        if not self.qa_pairs or self.qa_embeddings is None:
            return None

        query_embedding = self.embedding_model.encode([query])
        similarities = cosine_similarity(query_embedding, self.qa_embeddings)[0]

        best_idx = np.argmax(similarities)
        best_score = similarities[best_idx]

        if best_score > threshold:
            return {
                'qa': self.qa_pairs[best_idx],
                'score': float(best_score)  # Convert to Python float
            }

        return None

    def find_relevant_context(self, query, top_k=3, threshold=0.3):
        """Find most relevant contexts"""
        if not self.contexts or self.context_embeddings is None:
            return []

        query_embedding = self.embedding_model.encode([query])
        similarities = cosine_similarity(query_embedding, self.context_embeddings)[0]

        top_indices = np.argsort(similarities)[-top_k:][::-1]

        relevant_contexts = []
        for idx in top_indices:
            if similarities[idx] > threshold:
                relevant_contexts.append({
                    'context': self.contexts[idx],
                    'similarity': float(similarities[idx])  # Convert to Python float
                })

        return relevant_contexts

    def generate_qa_answer(self, question, context):
        """Generate answer using QA model"""
        if not self.qa_pipeline:
            return None

        try:
            # Truncate context if too long
            max_context_length = 500
            if len(context) > max_context_length:
                context = context[:max_context_length] + "..."

            result = self.qa_pipeline(question=question, context=context)

            if result['score'] > 0.15:  # Confidence threshold
                return result['answer']

        except Exception as e:
            print(f"QA Pipeline error: {e}")

        return None

    def get_smart_fallback(self, query):
        """Generate smart fallback responses"""
        query_lower = query.lower()

        # Keywords-based responses
        responses = {
            ('मंदिर',
             'temple'): "वाराणसी में काशी विश्वनाथ मंदिर, संकट मोचन हनुमान मंदिर, दुर्गा मंदिर जैसे प्रसिद्ध मंदिर हैं। किसी विशिष्ट मंदिर के बारे में पूछें।",
            ('घाट',
             'ghat'): "वाराणसी में दशाश्वमेध घाट, मणिकर्णिका घाट, अस्सी घाट जैसे प्रसिद्ध घाट हैं। किसी विशिष्ट घाट के बारे में जानना चाहते हैं?",
            ('आरती', 'aarti'): "गंगा आरती दशाश्वमेध घाट पर प्रतिदिन शाम को होती है। यह बहुत ही मनोहर और भव्य होती है।",
            ('गंगा', 'ganga'): "गंगा नदी वाराणसी की जीवनधारा है। यहाँ लोग स्नान करते हैं और आरती देखते हैं।",
            ('यात्रा', 'travel',
             'घूमना'): "वाराणसी में आप मंदिर, घाट, गलियाँ, और सांस्कृतिक स्थल देख सकते हैं। क्या विशिष्ट जानकारी चाहिए?"
        }

        for keywords, response in responses.items():
            if any(keyword in query_lower for keyword in keywords):
                return response

        return "मुझे वाराणसी के बारे में आपका प्रश्न समझ नहीं आया। कृपया मंदिर, घाट, आरती, या यात्रा के बारे में पूछें।"

    def process_query(self, query):
        """Main query processing function with evaluation metrics"""
        if not query.strip():
            return {
                'answer': "कृपया अपना प्रश्न पूछें।",
                'reference_answer': None,
                'evaluation_scores': None,
                'response_type': 'empty_query'
            }

        print(f"🔍 Processing query: {query}")

        # Step 1: Try to find direct Q&A match
        qa_match = self.find_best_qa_match(query)
        if qa_match:
            print(f"✅ Found Q&A match with score: {qa_match['score']:.3f}")
            
            # For direct Q&A matches, we have the reference answer
            predicted_answer = qa_match['qa']['answer']
            reference_answer = qa_match['qa']['answer']  # Same as predicted for exact matches
            
            # Calculate evaluation scores
            evaluation_scores = self.evaluation_metrics.calculate_all_scores(
                predicted_answer, reference_answer
            )
            
            return {
                'answer': predicted_answer,
                'reference_answer': reference_answer,
                'evaluation_scores': evaluation_scores,
                'response_type': 'direct_qa_match',
                'similarity_score': qa_match['score']
            }

        # Step 2: Find relevant contexts
        relevant_contexts = self.find_relevant_context(query)
        reference_answer = None
        
        if relevant_contexts:
            print(f"✅ Found {len(relevant_contexts)} relevant contexts")

            # Step 3: Try QA model on best context
            best_context = relevant_contexts[0]['context']
            qa_answer = self.generate_qa_answer(query, best_context['context'])

            if qa_answer:
                # Try to find a reference answer from the context's QAs
                reference_answer = self.find_reference_answer(query, best_context['qas'])
                
                evaluation_scores = self.evaluation_metrics.calculate_all_scores(
                    qa_answer, reference_answer
                ) if reference_answer else None
                
                return {
                    'answer': qa_answer,
                    'reference_answer': reference_answer,
                    'evaluation_scores': evaluation_scores,
                    'response_type': 'qa_model_generated',
                    'context_similarity': relevant_contexts[0]['similarity']
                }

            # Step 4: Check for direct Q&As in the context
            for qa in best_context['qas']:
                if self.is_similar_question(query, qa['question']):
                    predicted_answer = qa['answer']
                    reference_answer = qa['answer']
                    
                    evaluation_scores = self.evaluation_metrics.calculate_all_scores(
                        predicted_answer, reference_answer
                    )
                    
                    return {
                        'answer': predicted_answer,
                        'reference_answer': reference_answer,
                        'evaluation_scores': evaluation_scores,
                        'response_type': 'context_qa_match',
                        'context_similarity': relevant_contexts[0]['similarity']
                    }

        # Step 5: Smart fallback
        fallback_answer = self.get_smart_fallback(query)
        
        return {
            'answer': fallback_answer,
            'reference_answer': None,
            'evaluation_scores': None,
            'response_type': 'fallback'
        }

    def find_reference_answer(self, query, qas):
        """Find the most similar question's answer as reference"""
        if not qas:
            return None
        
        best_similarity = 0
        best_answer = None
        
        for qa in qas:
            if self.is_similar_question(query, qa['question'], threshold=0.5):
                # Calculate similarity score
                try:
                    embeddings = self.embedding_model.encode([query, qa['question']])
                    similarity = cosine_similarity([embeddings[0]], [embeddings[1]])[0][0]
                    
                    if similarity > best_similarity:
                        best_similarity = similarity
                        best_answer = qa['answer']
                except:
                    continue
        
        return best_answer

    def is_similar_question(self, q1, q2, threshold=0.7):
        """Check if two questions are similar"""
        try:
            embeddings = self.embedding_model.encode([q1, q2])
            similarity = cosine_similarity([embeddings[0]], [embeddings[1]])[0][0]
            return similarity > threshold
        except:
            return False


# Initialize improved chatbot
chatbot = ImprovedVATIKAChatbot()


def convert_to_serializable(obj):
    """Convert numpy types to Python native types for JSON serialization"""
    if isinstance(obj, np.floating):
        return float(obj)
    elif isinstance(obj, np.integer):
        return int(obj)
    elif isinstance(obj, np.ndarray):
        return obj.tolist()
    elif isinstance(obj, dict):
        return {key: convert_to_serializable(value) for key, value in obj.items()}
    elif isinstance(obj, list):
        return [convert_to_serializable(item) for item in obj]
    else:
        return obj


@app.route('/')
def home():
    return render_template('index.html')


@app.route('/chat', methods=['POST'])
def chat():
    try:
        data = request.get_json()
        user_message = data.get('message', '').strip()

        if not user_message:
            return jsonify({'error': 'कृपया कोई संदेश भेजें'}), 400

        # Process the query
        result = chatbot.process_query(user_message)

        # Convert all values to JSON serializable format
        result = convert_to_serializable(result)

        # Prepare response
        response_data = {
            'response': result['answer'],
            'status': 'success',
            'response_type': result['response_type'],
            'evaluation_scores': result['evaluation_scores'],
            'reference_answer': result['reference_answer'],
        }

        # Add additional info based on response type
        if result['response_type'] == 'direct_qa_match':
            response_data['similarity_score'] = result['similarity_score']
        elif result['response_type'] in ['qa_model_generated', 'context_qa_match']:
            response_data['context_similarity'] = result['context_similarity']

        # Add debug info in development
        if app.debug:
            response_data['debug'] = {
                'total_contexts': len(chatbot.contexts),
                'total_qas': len(chatbot.qa_pairs),
                'model_loaded': chatbot.qa_pipeline is not None
            }

        # Final conversion to ensure everything is serializable
        response_data = convert_to_serializable(response_data)

        return jsonify(response_data)

    except Exception as e:
        print(f"❌ Chat error: {e}")
        import traceback
        traceback.print_exc()
        return jsonify({
            'error': f'कुछ गलती हुई है: {str(e)}',
            'status': 'error'
        }), 500


@app.route('/health')
def health():
    return jsonify({
        'status': 'healthy',
        'contexts_loaded': len(chatbot.contexts),
        'qas_loaded': len(chatbot.qa_pairs),
        'embeddings_ready': chatbot.context_embeddings is not None,
        'qa_model_loaded': chatbot.qa_pipeline is not None
    })


@app.route('/debug')
def debug():
    """Debug endpoint to check data"""
    return jsonify({
        'contexts': len(chatbot.contexts),
        'qa_pairs': len(chatbot.qa_pairs),
        'sample_context': chatbot.contexts[0] if chatbot.contexts else None,
        'sample_qa': chatbot.qa_pairs[0] if chatbot.qa_pairs else None
    })
# Add this route AFTER your existing /chat route and BEFORE if __name__ == "__main__":

@app.route('/api2/predict', methods=['POST'])
def api_predict():
    """
    API endpoint for external applications
    Expected JSON: {"question": "your question", "context": "optional context"}
    Returns: {"answer": "response", "success": true, "evaluation_scores": {...}}
    """
    try:
        # Handle both JSON and form data
        if request.is_json:
            data = request.get_json()
        else:
            data = request.form.to_dict()
        
        # Get question from different possible keys
        question = data.get('question') or data.get('message') or data.get('query', '').strip()
        provided_context = data.get('context', '').strip()
        
        if not question:
            return jsonify({
                'success': False,
                'error': 'कृपया प्रश्न भेजें',
                'answer': ''
            }), 400
        
        # If context is provided, use it directly for QA
        if provided_context:
            # Use the provided context for QA
            if chatbot.qa_pipeline:
                try:
                    # Truncate context if too long
                    max_context_length = 500
                    if len(provided_context) > max_context_length:
                        provided_context = provided_context[:max_context_length] + "..."
                    
                    result = chatbot.qa_pipeline(question=question, context=provided_context)
                    
                    if result['score'] > 0.15:  # Confidence threshold
                        response_data = {
                            'success': True,
                            'answer': result['answer'],
                            'confidence': float(result['score']),
                            'response_type': 'context_based_qa',
                            'evaluation_scores': None,
                            'reference_answer': None
                        }
                        return jsonify(convert_to_serializable(response_data))
                except Exception as e:
                    print(f"Context QA error: {e}")
        
        # Use existing chatbot logic
        result = chatbot.process_query(question)
        
        # Convert to serializable format
        result = convert_to_serializable(result)
        
        # Prepare API response
        response_data = {
            'success': True,
            'answer': result['answer'],
            'response_type': result['response_type'],
            'evaluation_scores': result['evaluation_scores'],
            'reference_answer': result['reference_answer']
        }
        
        # Add additional info based on response type
        if result['response_type'] == 'direct_qa_match':
            response_data['similarity_score'] = result['similarity_score']
        elif result['response_type'] in ['qa_model_generated', 'context_qa_match']:
            response_data['context_similarity'] = result['context_similarity']
        
        return jsonify(response_data)
        
    except Exception as e:
        print(f"❌ API error: {e}")
        import traceback
        traceback.print_exc()
        return jsonify({
            'success': False,
            'error': f'कुछ गलती हुई है: {str(e)}',
            'answer': ''
        }), 500


@app.route('/api2/simple', methods=['POST', 'GET'])
def api_simple():
    """
    Simple API endpoint for basic question-answering
    POST: JSON {"question": "your question"}
    GET: ?question=your+question
    Returns: {"answer": "response"}
    """
    try:
        if request.method == 'POST':
            if request.is_json:
                data = request.get_json()
                question = data.get('question', '').strip()
            else:
                question = request.form.get('question', '').strip()
        else:  # GET
            question = request.args.get('question', '').strip()
        
        if not question:
            return jsonify({
                'answer': 'कृपया प्रश्न भेजें'
            }), 400
        
        # Process the query
        result = chatbot.process_query(question)
        
        # Simple response
        return jsonify({
            'answer': result['answer']
        })
        
    except Exception as e:
        print(f"❌ Simple API error: {e}")
        return jsonify({
            'answer': f'कुछ गलती हुई है: {str(e)}'
        }), 500

if __name__ == "__main__":
    # HF Spaces requirement: port 7860
    port = int(os.environ.get("PORT", 7860))
    app.run(host="0.0.0.0", port=port, debug=False)