Devansh959/Zomato-AI-Recommender

1import os2import sqlite33import pandas as pd4import json5from fastapi import FastAPI, HTTPException6from pydantic import BaseModel7from typing import Optional, List8from contextlib import asynccontextmanager9from dotenv import load_dotenv10from groq import Groq11 12load_dotenv()13 14GROQ_API_KEY = os.getenv("GROQ_API_KEY")15if GROQ_API_KEY:16    client = Groq(api_key=GROQ_API_KEY)17else:18    client = None19    print("WARNING: GROQ_API_KEY not found in .env file.")20 21df_restaurants = pd.DataFrame()22 23def get_clean_dataframe():24    db_path = os.path.join(os.path.dirname(os.path.dirname(__file__)), 'data', 'zomato.sqlite')25    with sqlite3.connect(db_path) as conn:26        df = pd.read_sql('SELECT * FROM restaurants', conn)27    28    df['clean_rate'] = df['rate'].astype(str).str.extract(r'(\d+\.\d+)').astype(float)29    df['clean_cost'] = df['approx_cost(for_two_people)'].astype(str).str.replace(',', '').str.extract(r'(\d+)').astype(float)30    df['clean_votes'] = pd.to_numeric(df['votes'], errors='coerce').fillna(0)31    32    return df33 34@asynccontextmanager35async def lifespan(app: FastAPI):36    global df_restaurants37    try:38        print("Loading and cleaning restaurant dataset into memory...")39        df_restaurants = get_clean_dataframe()40        print(f"Loaded {len(df_restaurants)} restaurants.")41    except Exception as e:42        print(f"Failed to load dataset: {e}")43    yield44    df_restaurants = pd.DataFrame()45 46app = FastAPI(title="Zomato Recommender API", lifespan=lifespan)47 48class UserPreferences(BaseModel):49    location: Optional[str] = None50    max_budget: Optional[float] = None51    cuisine: Optional[str] = None52    min_rating: Optional[float] = None53    additional_preferences: Optional[str] = None54 55@app.get("/api/locations")56def get_locations():57    if df_restaurants.empty:58        return {"locations": []}59    locs = df_restaurants['location'].dropna().unique().tolist()60    locs = sorted([str(loc).strip() for loc in locs if str(loc).strip()])61    return {"locations": ["Any"] + locs}62 63@app.post("/api/recommend")64def recommend_restaurants(prefs: UserPreferences):65    if df_restaurants.empty:66        raise HTTPException(status_code=500, detail="Dataset not loaded.")67    if not client:68        raise HTTPException(status_code=500, detail="Groq API key is missing. Check your .env file.")69        70    df = df_restaurants.copy()71    72    if prefs.location:73        loc_term = prefs.location.lower()74        df = df[df['location'].str.lower().str.contains(loc_term, na=False) | 75                df['listed_in(city)'].str.lower().str.contains(loc_term, na=False)]76    if prefs.cuisine:77        cuisine_term = prefs.cuisine.lower()78        df = df[df['cuisines'].str.lower().str.contains(cuisine_term, na=False)]79    if prefs.min_rating:80        df = df[df['clean_rate'] >= prefs.min_rating]81    if prefs.max_budget is not None:82        df = df[df['clean_cost'] <= prefs.max_budget]83            84    df = df.sort_values(by=['clean_rate', 'clean_votes'], ascending=[False, False])85    df = df.drop_duplicates(subset=['name'])86    87    # Get up to 15 matches to show to the user88    top_matches = df.head(15) 89    90    if len(top_matches) == 0:91        return {"recommendations": [], "message": "No restaurants found matching your exact criteria."}92    93    # Prepare MINIMAL JSON structure for Groq to save massive amounts of tokens94    context_data = []95    for _, row in top_matches.iterrows():96        context_data.append({97            "name": str(row['name']),98            "cuisines": str(row['cuisines']),99            "rating": row['clean_rate'] if pd.notnull(row['clean_rate']) else "Unknown"100        })101        102    system_prompt = """103    You are an expert Zomato restaurant recommendation AI. 104    You will receive a list of filtered restaurants and the user's exact preferences.105    Your task is to provide a personalized explanation for ALL the restaurants provided in the list.106    You MUST return ONLY a raw JSON object. Do NOT include markdown blocks like ```json.107    Format your response EXACTLY as this JSON object:108    {109      "recommendations": [110        {111          "restaurant_name": "Name",112          "ai_explanation": "A 1-2 sentence personalized explanation of why this fits the user's specific preferences."113        }114      ]115    }116    """117    118    user_prompt = f"""119    User Preferences:120    - Location: {prefs.location or 'Any'}121    - Max Budget for Two: {prefs.max_budget or 'Any'}122    - Cuisine: {prefs.cuisine or 'Any'}123    - Min Rating: {prefs.min_rating or 'Any'}124    - Additional Preferences: {prefs.additional_preferences or 'None specifically'}125    126    Here is the structured data of restaurants:127    {json.dumps(context_data, indent=2)}128    129    Please return the JSON object with explanations for all restaurants.130    """131    132    try:133        response = client.chat.completions.create(134            messages=[135                {"role": "system", "content": system_prompt},136                {"role": "user", "content": user_prompt}137            ],138            model="llama-3.1-8b-instant",139            temperature=0.2, 140            max_tokens=2000,141            response_format={"type": "json_object"}142        )143        144        ai_response_text = response.choices[0].message.content.strip()145        146        if ai_response_text.startswith("```json"):147            ai_response_text = ai_response_text[7:]148        if ai_response_text.endswith("```"):149            ai_response_text = ai_response_text[:-3]150            151        ai_json = json.loads(ai_response_text)152        ai_recs = ai_json.get("recommendations", [])153        154        # Map AI explanations back to the original full dataframe!155        exp_map = {r.get("restaurant_name"): r.get("ai_explanation") for r in ai_recs}156        157        final_results = []158        for _, row in top_matches.iterrows():159            name = str(row['name'])160            final_results.append({161                "restaurant_name": name,162                "cuisine": str(row['cuisines']),163                "rating": row['clean_rate'] if pd.notnull(row['clean_rate']) else None,164                "estimated_cost": str(row['approx_cost(for_two_people)']),165                "url": str(row['url']),166                "ai_explanation": exp_map.get(name, "A great match based on your standard filters.")167            })168            169        return {170            "total_matches_filtered": len(df),171            "recommendations": final_results172        }173    except Exception as e:174        raise HTTPException(status_code=500, detail=f"Error communicating with AI engine: {str(e)}")175