app.py

import pandas as pd
import numpy as np
from sentence_transformers import SentenceTransformer
import faiss
import gradio as gr
from typing import List, Dict, Tuple
import os
from huggingface_hub import InferenceClient

# STEP 1: Enhanced Restaurant Data (100+ items)
def create_enhanced_menu_data():
    """Create a larger, more diverse menu dataset"""
    menu_items = [
        # Healthy & Protein-Rich Options
        {"name": "Grilled Chicken Salad", "description": "High-protein grilled chicken breast with mixed greens, cherry tomatoes, cucumber. 35g protein, 450 calories", "category": "Salad", "cuisine": "Continental", "diet": "Non-Vegetarian", "price": 280},
        {"name": "Grilled Salmon with Veggies", "description": "Omega-3 rich grilled salmon fillet with steamed vegetables. Low-carb, high-protein meal. 40g protein", "category": "Main Course", "cuisine": "Continental", "diet": "Non-Vegetarian", "price": 480},
        {"name": "Protein Smoothie Bowl", "description": "Thick smoothie bowl with whey protein, banana, berries, topped with nuts and seeds. 30g protein", "category": "Breakfast", "cuisine": "Continental", "diet": "Vegetarian", "price": 250},
        {"name": "Tandoori Chicken Platter", "description": "Smoky grilled chicken marinated in yogurt and spices. High-protein, low-fat option. 45g protein", "category": "Starter", "cuisine": "Indian", "diet": "Non-Vegetarian", "price": 350},
        {"name": "Chicken Tikka Wrap", "description": "Grilled chicken tikka wrapped in whole wheat roti with veggies. Quick, portable, protein-packed meal. 30g protein", "category": "Snack", "cuisine": "Indian", "diet": "Non-Vegetarian", "price": 200},
        {"name": "Greek Yogurt Parfait", "description": "Layered Greek yogurt with fresh berries, granola, honey. High-protein breakfast with 20g protein, 300 calories", "category": "Breakfast", "cuisine": "Continental", "diet": "Vegetarian", "price": 180},
        {"name": "Egg White Omelette with Veggies", "description": "Fluffy egg white omelette loaded with vegetables. Low-calorie, high-protein breakfast. 20g protein, 150 calories", "category": "Breakfast", "cuisine": "Continental", "diet": "Non-Vegetarian", "price": 140},
        {"name": "Paneer Bhurji with Multigrain Toast", "description": "Scrambled cottage cheese with spices, served with multigrain toast. High-protein vegetarian breakfast. 25g protein", "category": "Breakfast", "cuisine": "Indian", "diet": "Vegetarian", "price": 160},
        
        # Comfort Food
        {"name": "Paneer Tikka Masala", "description": "Cottage cheese cubes in creamy tomato gravy with aromatic spices. Rich and indulgent comfort food perfect for cozy evenings", "category": "Main Course", "cuisine": "North Indian", "diet": "Vegetarian", "price": 290},
        {"name": "Butter Chicken with Naan", "description": "Tender chicken in rich, creamy tomato gravy. Indulgent comfort food perfect for cheat days", "category": "Main Course", "cuisine": "North Indian", "diet": "Non-Vegetarian", "price": 340},
        {"name": "Dal Tadka with Brown Rice", "description": "Yellow lentils tempered with cumin and garlic, served with brown rice. Light, healthy comfort food with balanced nutrition", "category": "Main Course", "cuisine": "North Indian", "diet": "Vegetarian", "price": 180},
        {"name": "Mac and Cheese", "description": "Creamy macaroni pasta with three types of cheese. Ultimate comfort food that soothes the soul", "category": "Main Course", "cuisine": "Continental", "diet": "Vegetarian", "price": 260},
        {"name": "Lentil Soup with Whole Wheat Bread", "description": "Hearty lentil soup with vegetables, served with whole wheat bread. Warming comfort food, high in fiber and protein", "category": "Soup", "cuisine": "Continental", "diet": "Vegetarian", "price": 160},
        {"name": "Spicy Szechuan Noodles", "description": "Fiery noodles tossed in Szechuan sauce with vegetables. Perfect when you're craving something spicy and bold", "category": "Main Course", "cuisine": "Chinese", "diet": "Vegetarian", "price": 220},
        
        # Vegan Options
        {"name": "Quinoa Buddha Bowl", "description": "Superfood bowl with quinoa, roasted vegetables, avocado, chickpeas. Vegan-friendly, high in fiber and nutrients. 15g protein, 400 calories", "category": "Salad", "cuisine": "Continental", "diet": "Vegan", "price": 320},
        {"name": "Vegan Pad Thai", "description": "Rice noodles stir-fried with tofu, peanuts, bean sprouts in tangy tamarind sauce. Plant-based and flavorful", "category": "Main Course", "cuisine": "Thai", "diet": "Vegan", "price": 300},
        {"name": "Vegan Chocolate Cake", "description": "Rich, moist chocolate cake made without dairy or eggs. Guilt-free indulgence for celebrations", "category": "Dessert", "cuisine": "Continental", "diet": "Vegan", "price": 180},
        {"name": "Hummus with Pita Bread", "description": "Creamy chickpea hummus served with warm pita bread. Protein-rich vegan appetizer", "category": "Starter", "cuisine": "Mediterranean", "diet": "Vegan", "price": 160},
        
        # Breakfast Items
        {"name": "Masala Dosa", "description": "Crispy rice crepe filled with spiced potato masala. Light yet filling breakfast option. Served with sambar and chutney", "category": "Breakfast", "cuisine": "South Indian", "diet": "Vegetarian", "price": 120},
        {"name": "Idli Sambar", "description": "Steamed rice cakes served with lentil soup and coconut chutney. Light and healthy breakfast", "category": "Breakfast", "cuisine": "South Indian", "diet": "Vegetarian", "price": 90},
        {"name": "Aloo Paratha", "description": "Whole wheat flatbread stuffed with spiced potato filling, served with curd. Hearty North Indian breakfast", "category": "Breakfast", "cuisine": "North Indian", "diet": "Vegetarian", "price": 100},
        {"name": "Avocado Toast with Eggs", "description": "Whole grain toast with mashed avocado, poached eggs, cherry tomatoes. Trendy breakfast loaded with healthy fats and protein", "category": "Breakfast", "cuisine": "Continental", "diet": "Vegetarian", "price": 240},
        {"name": "Pancakes with Maple Syrup", "description": "Fluffy pancakes served with butter and maple syrup. Classic comfort breakfast", "category": "Breakfast", "cuisine": "Continental", "diet": "Vegetarian", "price": 200},
        
        # Celebration Food
        {"name": "Chicken Biryani", "description": "Fragrant basmati rice with tender chicken, aromatic spices, served with raita. A hearty meal perfect for celebrations", "category": "Main Course", "cuisine": "Indian", "diet": "Non-Vegetarian", "price": 320},
        {"name": "Mutton Rogan Josh", "description": "Tender mutton cooked in aromatic Kashmiri spices. A royal dish perfect for special occasions", "category": "Main Course", "cuisine": "North Indian", "diet": "Non-Vegetarian", "price": 420},
        {"name": "Lobster Thermidor", "description": "Luxurious lobster in creamy sauce, perfect for celebrations and special moments", "category": "Main Course", "cuisine": "French", "diet": "Non-Vegetarian", "price": 980},
        
        # More variety
        {"name": "Margherita Pizza", "description": "Classic pizza with fresh mozzarella, tomato sauce, and basil leaves", "category": "Main Course", "cuisine": "Italian", "diet": "Vegetarian", "price": 350},
        {"name": "Chicken Wings", "description": "Crispy fried chicken wings tossed in spicy buffalo sauce", "category": "Starter", "cuisine": "Continental", "diet": "Non-Vegetarian", "price": 280},
        {"name": "Caesar Salad", "description": "Fresh romaine lettuce, croutons, parmesan cheese, caesar dressing", "category": "Salad", "cuisine": "Continental", "diet": "Vegetarian", "price": 220},
        {"name": "Tom Yum Soup", "description": "Hot and sour Thai soup with mushrooms, lemongrass, lime. Light, flavorful, perfect when feeling under the weather", "category": "Soup", "cuisine": "Thai", "diet": "Vegetarian", "price": 180},
        {"name": "Fish and Chips", "description": "Crispy battered fish fillet with french fries and tartar sauce", "category": "Main Course", "cuisine": "Continental", "diet": "Non-Vegetarian", "price": 380},
        {"name": "Paneer Tikka", "description": "Marinated cottage cheese chunks grilled with bell peppers and onions", "category": "Starter", "cuisine": "Indian", "diet": "Vegetarian", "price": 240},
        {"name": "Veg Spring Rolls", "description": "Crispy rolls filled with mixed vegetables and served with sweet chili sauce", "category": "Starter", "cuisine": "Chinese", "diet": "Vegetarian", "price": 140},
        {"name": "Chocolate Brownie with Ice Cream", "description": "Rich chocolate brownie with vanilla ice cream. Perfect dessert for celebrations", "category": "Dessert", "cuisine": "Continental", "diet": "Vegetarian", "price": 160},
        {"name": "Gulab Jamun", "description": "Soft milk dumplings soaked in sugar syrup. Traditional Indian sweet perfect for festivals", "category": "Dessert", "cuisine": "Indian", "diet": "Vegetarian", "price": 80},
        {"name": "Tiramisu", "description": "Italian dessert with coffee-soaked ladyfingers and mascarpone cream", "category": "Dessert", "cuisine": "Italian", "diet": "Vegetarian", "price": 220},
        {"name": "Sushi Platter", "description": "Assorted sushi rolls with wasabi and soy sauce. Light and healthy Japanese delicacy", "category": "Main Course", "cuisine": "Japanese", "diet": "Non-Vegetarian", "price": 680},
        {"name": "Ramen Bowl", "description": "Japanese noodle soup with rich broth, vegetables, and choice of protein. Comforting and flavorful", "category": "Main Course", "cuisine": "Japanese", "diet": "Non-Vegetarian", "price": 380},
        {"name": "Falafel Wrap", "description": "Crispy chickpea falafel in pita with tahini sauce and fresh veggies. Healthy vegan option", "category": "Snack", "cuisine": "Mediterranean", "diet": "Vegan", "price": 180},
    ]
    
    return pd.DataFrame(menu_items)

def load_restaurant_data():
    """Try loading from Kaggle, fallback to enhanced local data"""
    try:
        import kagglehub
        print("Attempting to download from Kaggle...")
        path = kagglehub.dataset_download("graphquest/restaurant-menu-items")
        df = pd.read_csv(f"{path}/menu_items.csv")
        df = df.dropna(subset=['name', 'description'])
        
        # Ensure required columns
        if 'category' not in df.columns:
            df['category'] = 'Main Course'
        if 'price' not in df.columns:
            df['price'] = np.random.randint(100, 500, len(df))
        if 'cuisine' not in df.columns:
            df['cuisine'] = 'International'
        if 'diet' not in df.columns:
            df['diet'] = 'Vegetarian'
            
        print(f"✅ Loaded {len(df)} items from Kaggle!")
        return df.head(500)
        
    except Exception as e:
        print(f"Kaggle download failed, using local enhanced dataset with 37 items")
        return create_enhanced_menu_data()

# STEP 2: Real LLM using Hugging Face
class HuggingFaceReasoningGenerator:
    """Uses Hugging Face Inference API for real LLM generation"""
    
    def __init__(self, hf_token=None):
        self.hf_token = hf_token or os.getenv("HF_TOKEN")
        
        if self.hf_token:
            # Using a good open-source model
            self.client = InferenceClient(token=self.hf_token)
            self.model = "mistralai/Mistral-7B-Instruct-v0.2"
            print(f"✅ Using real LLM: {self.model}")
        else:
            print("⚠️ No HF token provided, will use fallback reasoning")
            self.client = None
    
    def generate_reasoning(self, dish_name: str, dish_description: str, 
                          user_query: str, context: Dict) -> str:
        """Generate personalized reasoning using Hugging Face LLM"""
        
        if not self.client:
            return self._fallback_reasoning(dish_name, dish_description, context)
        
        prompt = f"""You are a food recommendation assistant. Be concise and friendly.

User Query: "{user_query}"
Dish: {dish_name}
Description: {dish_description}
Context: Situation: {', '.join(context.get('situation', []))}, Dietary: {', '.join(context.get('dietary', []))}

Write ONE short sentence (max 20 words) explaining why this dish is perfect for the user. Be specific and natural.

Reasoning:"""

        try:
            response = self.client.text_generation(
                prompt,
                model=self.model,
                max_new_tokens=50,
                temperature=0.7,
                return_full_text=False
            )
            
            # Clean up the response
            reasoning = response.strip()
            if reasoning:
                reasoning = reasoning.split('.')[0] + '.'
                return reasoning
            else:
                return self._fallback_reasoning(dish_name, dish_description, context)
                
        except Exception as e:
            print(f"LLM Error: {e}")
            return self._fallback_reasoning(dish_name, dish_description, context)
    
    def _fallback_reasoning(self, dish_name, dish_description, context):
        """Fallback when LLM fails"""
        desc_lower = dish_description.lower()
        
        if 'post_workout' in context.get('situation', []) and 'protein' in desc_lower:
            return f"Excellent post-workout choice with high protein for muscle recovery."
        
        if 'celebration' in context.get('situation', []):
            return f"Perfect for celebrations with its indulgent and special flavors."
        
        if 'comfort' in context.get('situation', []):
            return f"Comforting and satisfying when you need a mood boost."
        
        if 'vegan' in context.get('dietary', []):
            return f"Great plant-based option that's both nutritious and delicious."
        
        return f"This dish matches your preferences perfectly."

# STEP 3: Context Extractor
class ContextExtractor:
    def __init__(self):
        self.context_patterns = {
            'post_workout': ['workout', 'gym', 'exercise', 'training', 'fitness'],
            'celebration': ['celebration', 'party', 'birthday', 'anniversary', 'special'],
            'comfort': ['stressed', 'tired', 'comfort', 'cozy', 'relax', 'bad day'],
            'healthy': ['healthy', 'diet', 'nutrition', 'fit', 'wellness'],
        }
        
        self.dietary_patterns = {
            'vegetarian': ['vegetarian', 'veg', 'no meat'],
            'vegan': ['vegan', 'plant-based'],
            'protein': ['protein', 'high protein'],
        }
    
    def extract_context(self, query: str) -> Dict:
        query_lower = query.lower()
        context = {'situation': [], 'dietary': []}
        
        for situation, keywords in self.context_patterns.items():
            if any(kw in query_lower for kw in keywords):
                context['situation'].append(situation)
        
        for diet, keywords in self.dietary_patterns.items():
            if any(kw in query_lower for kw in keywords):
                context['dietary'].append(diet)
        
        return context

# STEP 4: Neural Search System
class GenAINeuralSearch:
    def __init__(self, menu_df: pd.DataFrame, hf_token=None):
        print("🚀 Initializing GenAI Neural Search System...")
        self.model = SentenceTransformer('all-MiniLM-L6-v2')
        self.menu_df = menu_df
        self.context_extractor = ContextExtractor()
        self.llm_generator = HuggingFaceReasoningGenerator(hf_token)
        
        self.menu_df['search_text'] = self.menu_df.apply(
            lambda row: f"{row['name']}. {row['description']}. {row['category']}. {row['cuisine']}. {row['diet']}.",
            axis=1
        )
        
        print("📊 Creating embeddings...")
        self.embeddings = self.model.encode(
            self.menu_df['search_text'].tolist(),
            show_progress_bar=True
        )
        
        print("🔍 Building FAISS index...")
        dimension = self.embeddings.shape[1]
        self.index = faiss.IndexFlatIP(dimension)
        faiss.normalize_L2(self.embeddings)
        self.index.add(self.embeddings)
        
        print(f"✅ System ready with {len(self.menu_df)} menu items!")
    
    def search_with_genai(self, query: str, top_k: int = 5) -> Tuple[List[Dict], Dict]:
        context = self.context_extractor.extract_context(query)
        
        query_embedding = self.model.encode([query])
        faiss.normalize_L2(query_embedding)
        similarities, indices = self.index.search(query_embedding, top_k)
        
        results = []
        for idx, score in zip(indices[0], similarities[0]):
            item = self.menu_df.iloc[idx]
            
            # Generate LLM reasoning
            llm_reasoning = self.llm_generator.generate_reasoning(
                dish_name=item['name'],
                dish_description=item['description'],
                user_query=query,
                context=context
            )
            
            results.append({
                'name': item['name'],
                'description': item['description'],
                'category': item['category'],
                'cuisine': item['cuisine'],
                'diet': item['diet'],
                'price': f"₹{item['price']}",
                'similarity_score': float(score),
                'ai_reasoning': llm_reasoning
            })
        
        return results, context

# STEP 5: Initialize
print("Loading menu data...")
menu_df = load_restaurant_data()
print("Initializing search system...")
search_system = GenAINeuralSearch(menu_df, hf_token=os.getenv("HF_TOKEN"))

# STEP 6: Gradio Interface
def search_with_genai(query, num_results):
    if not query.strip():
        return "<p style='color: #666;'>Please enter a search query to get started!</p>"
    
    results, context = search_system.search_with_genai(query, top_k=num_results)
    
    # Context display
    html_output = "<div style='background: linear-gradient(135deg, #667eea 0%, #764ba2 100%); padding: 20px; border-radius: 12px; margin-bottom: 25px; color: white;'>"
    html_output += f"<h3 style='margin: 0 0 15px 0; font-size: 24px;'>🔍 Results for: \"{query}\"</h3>"
    
    if context['situation'] or context['dietary']:
        html_output += "<div style='background: rgba(255,255,255,0.2); padding: 12px; border-radius: 8px;'>"
        html_output += "<strong>🧠 Context Understanding:</strong><br/>"
        if context['situation']:
            html_output += f"<span style='margin-right: 10px;'>📍 Situation: {', '.join(context['situation'])}</span>"
        if context['dietary']:
            html_output += f"<span>🥗 Dietary: {', '.join(context['dietary'])}</span>"
        html_output += "</div>"
    
    html_output += "</div>"
    
    # Results
    for i, result in enumerate(results, 1):
        html_output += f"""
        <div style='background: white; border: 2px solid #e0e0e0; padding: 20px; margin: 20px 0; border-radius: 12px; box-shadow: 0 2px 8px rgba(0,0,0,0.1);'>
            <div style='display: flex; justify-content: space-between; align-items: center; margin-bottom: 15px;'>
                <h4 style='margin: 0; color: #2c3e50; font-size: 20px;'>{i}. {result['name']}</h4>
                <span style='color: #27ae60; font-weight: bold; font-size: 18px;'>{result['price']}</span>
            </div>
            
            <div style='background: #f0f9ff; border-left: 4px solid #3b82f6; padding: 12px; margin: 15px 0; border-radius: 6px;'>
                <strong style='color: #1e40af;'>🤖 AI Reasoning:</strong>
                <p style='margin: 5px 0 0 0; color: #374151;'>{result['ai_reasoning']}</p>
            </div>
            
            <p style='color: #555; line-height: 1.6; margin: 15px 0;'>{result['description']}</p>
            
            <div style='margin-top: 15px;'>
                <span style='background: #3b82f6; color: white; padding: 6px 12px; border-radius: 20px; font-size: 13px; margin-right: 8px;'>{result['category']}</span>
                <span style='background: #ef4444; color: white; padding: 6px 12px; border-radius: 20px; font-size: 13px; margin-right: 8px;'>{result['cuisine']}</span>
                <span style='background: #10b981; color: white; padding: 6px 12px; border-radius: 20px; font-size: 13px;'>{result['diet']}</span>
            </div>
            
            <p style='color: #9ca3af; font-size: 12px; margin: 10px 0 0 0;'>Relevance: {result['similarity_score']:.1%}</p>
        </div>
        """
    
    return html_output

# Gradio Interface
with gr.Blocks(theme=gr.themes.Soft(), title="Swiggy Neural Search - GenAI") as demo:
    gr.Markdown("""
    # 🍽️ Swiggy Neural Search - GenAI Edition
    
    **Powered by:** Sentence Transformers + FAISS + Hugging Face LLM (Mistral-7B)
    
    This system understands your context and generates personalized explanations using AI!
    """)
    
    with gr.Row():
        with gr.Column(scale=2):
            query_input = gr.Textbox(
                label="What are you looking for?",
                placeholder="Try: 'I just finished my workout. Show me healthy lunch options'",
                lines=3
            )
            num_results = gr.Slider(minimum=3, maximum=8, value=5, step=1, label="Number of results")
            search_btn = gr.Button("🔍 Search", variant="primary", size="lg")
        
    output_html = gr.HTML(label="Results")
    
    gr.Examples(
        examples=[
            ["I just finished my workout. Show me healthy lunch options"],
            ["I'm feeling stressed, need comfort food"],
            ["It's my birthday! Show me celebration-worthy food"],
            ["Vegan protein-rich dinner options"],
            ["Something light for breakfast"],
        ],
        inputs=query_input
    )
    
    search_btn.click(
        fn=search_with_genai,
        inputs=[query_input, num_results],
        outputs=output_html
    )

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