Update app.py

app.py

@@ -12,13 +12,11 @@ from pathlib import Path
 import torch
 import torch.nn.functional as F
 
-# from src.data.embs import ImageDataset
 from src.model.blip_embs import blip_embs
 from src.data.transforms import transform_test
 
-from transformers import StoppingCriteria, StoppingCriteriaList, TextIteratorStreamer
+from transformers import StoppingCriteria, StoppingCriteriaList
 import gradio as gr
-# import spaces
 
 from langchain.chains import ConversationChain
 from langchain_community.chat_message_histories import ChatMessageHistory
@@ -35,7 +33,6 @@ from flask_socketio import SocketIO, emit
 import json
 from openai import OpenAI
 
-# GROQ_API_KEY = os.getenv("GROQ_API_KEY")
 load_dotenv(".env")
 USER_AGENT = os.getenv("USER_AGENT")
 GROQ_API_KEY = os.getenv("GROQ_API_KEY")
@@ -55,7 +52,6 @@ app.config['MAX_CONTENT_LENGTH'] = 1024 * 1024 * 1024
 socketio = SocketIO(app, cors_allowed_origins="*", logger=True, max_http_buffer_size=1024 * 1024 * 1024)
 app.config['SECRET_KEY'] = SECRET_KEY
 
-
 # Initialize LLM
 llm = ChatGroq(model="llama-3.1-8b-instant", temperature=0, max_tokens=1024, max_retries=2)
 
@@ -65,7 +61,7 @@ json_llm = ChatGroq(model="llama-3.1-70b-versatile", temperature=0, max_tokens=1
 # Initialize Router
 router = ChatGroq(model="llama-3.2-3b-preview", temperature=0, max_tokens=1024, max_retries=2, model_kwargs={"response_format": {"type": "json_object"}})
 
-# Initialize Router
+# Initialize answer formatter
 answer_formatter = ChatGroq(model="llama-3.1-8b-instant", temperature=0, max_tokens=1024, max_retries=2)
 
 # Initialized recommendation LLM
@@ -180,66 +176,30 @@ class Chat:
     def ask(self):
         return json.dumps(self.target_recipe.to_json())
 
-
-
 chat = Chat(model,transform,df,tar_img_feats, device)
 
-
 def answer_generator(formated_input, session_id):
     # QA system prompt and chain
     qa_system_prompt = """
-    You are an AI assistant developed by Nutrigenics AI, specializing in intelligent recipe information retrieval and recipe suggestions. Your purpose is to help users by recommending recipes, providing detailed nutritional values, listing ingredients, offering step-by-step cooking instructions, and filtering recipes based on context and user queries.
-    Operational Guidelines: \n
-    1. Input Structure: \n
-    - Context: You may receive contextual information related to recipes, such as specific recipe name, ingredients, nutritional informations, intsructions, recipe tags, or previously selected dishes. \n
-    - User Query: Users will pose questions or requests related to recipes, nutritional information, ingredient, cooking instructions, and more. \n
-    2. Response Strategy: \n
-    - Utilize Provided Context: If the context contains relevant information that addresses the user's query, base your response on this provided data to ensure accuracy and relevance. \n
-    - Respond to User Query Directly: If the context does not contain the necessary information to answer the user's query, kindly state that you do not have the required information. \n
-    Output Format: \n
-    - The output format should be JSON.
-    - The output should have a key 'header' with response message header such as "Here is your ....",
-    - Then there should be other key with the actual response information. If the user query asks recipe ingredients then the key should be named "ingredients" with
-    JSON object as its value. The JSON object should have ingredient and its measurement as key-value pairs. Similarly if user asked for nutritional information then the output should have 'header' key with header text and 'nutrients' key
-    with a JSON object og nutrient and its content as key-value pairs. Similarly if the user query asks for recipe instructions then JSON output should include 'header key with header text and 
-    'instructions' key with a list of instructions as its value.
-    Following are the output formats for some cases:
-        1. if user query asks for all recipe information, then output should be of following format:
-            {
-                header: header text,
-                recipe_name: Recipe Name,
-                recipe_instructions: List of recipe instructions,
-                recipe_nutrients: key-value pairs of nutrients name and its content,
-                recipe_ingredients: key-value pairs of ingredients name and its content,
-                recipe_tags: List of tags related to recipe,
-                .
-                .
-                .
-            }
-    
-        2. if user query asks for recipe nutrients information, then output should be of following format:
-            {
-                header: header text,
-                recipe_nutrients: key-value pairs of nutrients name and its content.
-            }
-        
-        3. if user query asks for recipe instructions information, then output should be of following format:
-            {
-                header: header text,
-                recipe_instructions: List of recipe instructions,
-            }
-        4. if user query asks for recipe instructions information, then output should be of following format:
-            {
-                header: header text,
-                recipe_instructions: List of recipe instructions,
-            }
+    You are an AI assistant developed by Nutrigenics AI. Your purpose is to help users by providing accurate and relevant answers to their questions.
+    Operational Guidelines:
+
+    1. Input Structure:
+    - Context: You may receive contextual information related to recipes or other topics.
+    - User Query: Users will pose questions or requests on various topics.
+
+    2. Response Strategy:
+    - Utilize Provided Context: If the context contains relevant information that addresses the user's query, base your response on this provided data.
+    - Respond to User Query Directly: If the context does not contain the necessary information, answer the question to the best of your ability.
+
+    Output Format:
+    - Provide clear and concise answers.
+    - Format your response in JSON with a key 'content' containing your answer.
+
     Additional Instructions:
-    - Precision and Personalization: Always aim to provide precise, personalized, and relevant information to users based on both the provided context and their specific queries.
-    - Clarity and Coherence: Ensure all responses are clear, well-structured, and easy to understand, facilitating a seamless user experience.
-    - Substitute Suggestions: Consider user preferences and dietary restrictions outlined in the context or user query when suggesting ingredient substitutes.
-    - Dynamic Adaptation: Adapt your responses dynamically based on whether the context is relevant to the user's current request, ensuring optimal use of available information.
-    - Don't mention about the context in the response, format the answer in a natural and friendly way.
-    
+    - Precision and Personalization: Always aim to provide precise, personalized, and relevant information.
+    - Clarity and Coherence: Ensure all responses are clear, well-structured, and easy to understand.
+    - Do not mention about the context in the response, format the answer in a natural and friendly way.
     """
     qa_prompt = ChatPromptTemplate.from_messages(
         [
@@ -263,119 +223,26 @@ def answer_generator(formated_input, session_id):
 
     return response
 
-
 def json_answer_generator(user_query, context):
     system_prompt = """
-            Given a recipe context in JSON format, respond to user queries by extracting and returning the requested information in JSON format with an additional `"header"` key containing a response starter. Use the following rules:
-            1. **Recipe Information Extraction**: 
-            - If the user query explicitly requests specific recipe data (e.g., ingredients, nutrients, or instructions), return only those JSON objects from the provided recipe context.
-            - For example, if the user asks, “What are the ingredients?” or “Show me the nutrient details,” your output should be limited to only the requested JSON objects (e.g., `recipe_ingredients`, `recipe_nutrients`).
-            - Include `"header": "Here is the information you requested:"` at the start of each response.
-            2. **Multiple Information Points**:
-            - If a user query asks for more than one piece of information, return each requested JSON object from the recipe context in a combined JSON response.
-            - For example, if the query is “Give me the ingredients and instructions,” the output should include both `recipe_ingredients` and `recipe_instructions` objects.
-            - Include `"header": "Here is the information you requested:"` at the start of each response.
-            3. **Non-Specific Recipe Information**:
-            - If the query does not directly refer to recipe data but instead asks for a general response based on the context, return a JSON object with a single key `"content"` and a descriptive response as its value.
-            - Include `"header": "Here is a suggestion based on the recipe:"` as the response starter.
-            - For example, if the query is “How can I use this recipe for a healthy lunch?” return a response like:
-                ```json
-                {
-                    "header": "Here is a suggestion based on the recipe:",
-                    "content": "This Asian Potato Salad with Seven Minute Egg is a nutritious and light option, ideal for a balanced lunch. It provides protein and essential nutrients with low calories."
-                }
-                ```
-            **Example Context**:
-            ```json
-            {
-                "recipe_name": "Asian Potato Salad with Seven Minute Egg",
-                "recipe_time": 0,
-                "recipe_yields": "4 servings",
-                "recipe_ingredients": [
-                    "2 1/2 cup Multi-Colored Fingerling Potato",
-                    "3/4 cup Celery",
-                    "1/4 cup Red Onion",
-                    "2 tablespoon Fresh Parsley",
-                    "1/3 cup Mayonnaise",
-                    "1 tablespoon Chili Garlic Sauce",
-                    "1 teaspoon Hoisin Sauce",
-                    "1 splash Soy Sauce",
-                    "to taste Salt",
-                    "to taste Ground Black Pepper",
-                    "4 Egg"
-                ],
-                "recipe_instructions": "Fill a large stock pot with water. Add the Multi-Colored Fingerling Potato...",
-                "recipe_image": "https://www.sidechef.com/recipe/eeeeeceb-493e-493d-8273-66c800821b13.jpg?d=1408x1120",
-                "blogger": "sidechef.com",
-                "recipe_nutrients": {
-                    "calories": "80 calories",
-                    "proteinContent": "2.1 g",
-                    "fatContent": "6.2 g",
-                    "carbohydrateContent": "3.9 g",
-                    "fiberContent": "0.5 g",
-                    "sugarContent": "0.4 g",
-                    "sodiumContent": "108.0 mg",
-                    "saturatedFatContent": "1.2 g",
-                    "transFatContent": "0.0 g",
-                    "cholesterolContent": "47.4 mg",
-                    "unsaturatedFatContent": "3.8 g"
-                },
-                "tags": [
-                    "Salad",
-                    "Lunch",
-                    "Brunch",
-                    "Appetizers",
-                    "Side Dish",
-                    "Budget-Friendly",
-                    "Vegetarian",
-                    "Pescatarian",
-                    "Eggs",
-                    "Potatoes",
-                    "Easy",
-                    "Dairy-Free",
-                    "Shellfish-Free",
-                    "Entertaining",
-                    "Fish-Free",
-                    "Peanut-Free",
-                    "Tree Nut-Free",
-                    "Sugar-Free",
-                    "Global",
-                    "Tomato-Free",
-                    "Stove",
-                    ""
-                ],
-                "id_": "0000001"
-            }
-            **Example Query & Output**:
-            **Query**: "What are the ingredients and calories?"
-            **Output**:
-            ```json
-            {
-                "header": "Here is the information you requested:",
-                "recipe_ingredients": [
-                    "2 1/2 cup Multi-Colored Fingerling Potato",
-                    "3/4 cup Celery",
-                    "1/4 cup Red Onion",
-                    "2 tablespoon Fresh Parsley",
-                    "1/3 cup Mayonnaise",
-                    "1 tablespoon Chili Garlic Sauce",
-                    "1 teaspoon Hoisin Sauce",
-                    "1 splash Soy Sauce",
-                    "to taste Salt",
-                    "to taste Ground Black Pepper",
-                    "4 Egg"
-                ],
-                "recipe_nutrients": {
-                    "calories": "80 calories"
-                }
-            }
-        Try to format the output as JSON object with key value pairs.
+    Given a context in JSON format, respond to user queries by extracting and returning the requested information in JSON format with an additional `"header"` key containing a response starter. Use the following rules:
+
+    1. **Information Extraction**: 
+    - If the user query explicitly requests specific data (e.g., ingredients, nutrients, or instructions), return only those JSON objects from the provided context.
+    - Include `"header": "Here is the information you requested:"` at the start of each response.
+
+    2. **General Responses**:
+    - If the query is not directly related to the context, provide a helpful and accurate answer.
+    - Include `"header": "Here is your answer:"` at the start of the response.
+    - Return a JSON object with a single key `"content"` and your response as its value.
+
+    Try to format the output as a JSON object with key-value pairs.
     """
 
     formatted_input = f"""
-        User Query: {user_query}
-        Recipe data as Context:
-        {context}
+    User Query: {user_query}
+    Context:
+    {context}
     """
     response = json_llm.invoke(
         [SystemMessage(content=system_prompt)]
@@ -388,19 +255,13 @@ def json_answer_generator(user_query, context):
     res = json.loads(response.content)
     return res
 
-
-
-
-### Router
-import json
-from langchain_core.messages import HumanMessage, SystemMessage
-
 def router_node(query):
     # Prompt
-    router_instructions = """You are an expert at determining the appropriate task for a user’s question based on chat history and the current query context. You have two available tasks:
-        1.	Retrieval: Fetch information based on user's chat history and current query.
-        2.	Recommendation/Suggestion: Recommend recipes to users based on the query.
-    Return a JSON response with a single key named “task” indicating either “retrieval” or “recommendation” based on your decision.
+    router_instructions = """You are an expert at determining the appropriate task for a user’s question based on chat history and the current query context. You have three available tasks:
+        1. Retrieval: Fetch information based on user's chat history and current query.
+        2. Recommendation/Suggestion: Recommend recipes to users based on the query.
+        3. General: Answer general questions not related to recipes or the current context.
+    Return a JSON response with a single key named “task” indicating either “retrieval”, “recommendation”, or “general” based on your decision.
     """
     response = router.invoke(
         [SystemMessage(content=router_instructions)]
@@ -415,9 +276,11 @@ def router_node(query):
 
 def recommendation_node(query):
     prompt = """
-    You are a helpful assistant that writes Python code to filter recipes from a JSON filr based o the user query. \n
-    JSON file path = 'recipes.json' \n
-    The JSON file is a list of recipes with the following structure: \n
+    You are a helpful assistant that writes Python code to filter recipes from a JSON file based on the user query.
+
+    JSON file path = 'recipes.json'
+
+    The JSON file is a list of recipes with the following structure:
     {
         "recipe_name": string,
         "recipe_time": integer,
@@ -428,72 +291,25 @@ def recommendation_node(query):
         "blogger": string,
         "recipe_nutrients": JSON object with key-value pairs such as "protein: 10g",
         "tags": list of tags related to a recipe
-    } \n
-    Here is the example of a recipe json object from the JSON data: \n
-    {
-        "recipe_name": "Asian Potato Salad with Seven Minute Egg",
-        "recipe_time": 0,
-        "recipe_yields": "4 servings",
-        "recipe_ingredients": [
-            "2 1/2 cup Multi-Colored Fingerling Potato",
-            "3/4 cup Celery",
-            "1/4 cup Red Onion",
-            "2 tablespoon Fresh Parsley",
-            "1/3 cup Mayonnaise",
-            "1 tablespoon Chili Garlic Sauce",
-            "1 teaspoon Hoisin Sauce",
-            "1 splash Soy Sauce",
-            "to taste Salt",
-            "to taste Ground Black Pepper",
-            "4 Egg"
-        ],
-        "recipe_instructions": "Fill a large stock pot with water.\nAdd the Multi-Colored Fingerling Potato (2 1/2 cup) and bring water to a boil. Boil the potatoes for 20 minutes or until fork tender.\nDrain the potatoes and let them cool completely.\nMeanwhile, mix together in a small bowl Mayonnaise (1/3 cup), Chili Garlic Sauce (1 tablespoon), Hoisin Sauce (1 teaspoon), and Soy Sauce (1 splash).\nTo make the Egg (4), fill a stock pot with water and bring to a boil Gently add the eggs to the water and set a timer for seven minutes.\nThen move the eggs to an ice bath to cool completely. Once cooled, crack the egg slightly and remove the shell. Slice in half when ready to serve.\nNext, halve the cooled potatoes and place into a large serving bowl. Add the Ground Black Pepper (to taste), Celery (3/4 cup), Red Onion (1/4 cup), and mayo mixture. Toss to combine adding Salt (to taste) and Fresh Parsley (2 tablespoon).\nTop with seven minute eggs and serve. Enjoy!",
-        "recipe_image": "https://www.sidechef.com/recipe/eeeeeceb-493e-493d-8273-66c800821b13.jpg?d=1408x1120",
-        "blogger": "sidechef.com",
-        "recipe_nutrients": {
-            "calories": "80 calories",
-            "proteinContent": "2.1 g",
-            "fatContent": "6.2 g",
-            "carbohydrateContent": "3.9 g",
-            "fiberContent": "0.5 g",
-            "sugarContent": "0.4 g",
-            "sodiumContent": "108.0 mg",
-            "saturatedFatContent": "1.2 g",
-            "transFatContent": "0.0 g",
-            "cholesterolContent": "47.4 mg",
-            "unsaturatedFatContent": "3.8 g"
-        },
-        "tags": [
-            "Salad",
-            "Lunch",
-            "Brunch",
-            "Appetizers",
-            "Side Dish",
-            "Budget-Friendly",
-            "Vegetarian",
-            "Pescatarian",
-            "Eggs",
-            "Potatoes",
-            "Dairy-Free",
-            "Shellfish-Free"
-        ]
-    } \n
-    Based on the user query, provide a Python function to filter the JSON data. The output of the function should be a list of json objects. \n
+    }
+
+    Based on the user query, provide a Python function to filter the JSON data. The output of the function should be a list of JSON objects.
+
     Recipe filtering instructions:
     - If a user asked for the highest nutrient recipe such as "high protein or high calories" then filtered recipes should be the top highest recipes from all the recipes with high nutrient.
-    - sort or rearrange recipes based which recipes are more appropriate for the user.
+    - Sort or rearrange recipes based on which recipes are more appropriate for the user.
+
     Your output instructions:
     - The function name should be filter_recipes. The input to the function should be file name.
     - The length of output recipes should not be more than 6.
-    - Only give me output function. Do not call the function.
-    - Give the python function as a key named "code" in a json format.
+    - Only give me the output function. Do not call the function.
+    - Give the python function as a key named "code" in a JSON format.
     - Do not include any other text with the output, only give python code.
     - If you do not follow the above given instructions, the chat may be terminated.
     """
     max_tries = 3
     while True:
         try:
-            # llm = ChatGroq(model="llama-3.1-8b-instant", temperature=0, max_tokens=1024, max_retries=2, model_kwargs={"response_format": {"type": "json_object"}})
             response = client.chat.completions.create(
                 model="gpt-4o-mini",
                 messages=[
@@ -521,10 +337,9 @@ def recommendation_node(query):
                 max_tries -= 1
     return filtered_recipes
 
-
 def answer_formatter_node(question, context):
-    prompt = f""" You are an highly clever question-answering assistant trained to provide clear and concise answers based on the user query and provided context. 
-    Your task is to generated answers for the user query based on the context provided.
+    prompt = f"""You are a highly clever question-answering assistant trained to provide clear and concise answers based on the user query and provided context. 
+    Your task is to generate answers for the user query based on the context provided.
     Instructions for your response:
     1. Directly answer the user query using only the information provided in the context.
     2. Ensure your response is clear and concise.
@@ -543,19 +358,16 @@ def answer_formatter_node(question, context):
     return res
 
 def reguar_answer_node(question, context):
-    prompt = f""" You are an highly clever question-answering assistant trained to provide clear and concise answers based on the user query and provided context. 
-    Your task is to generated answers for the user query based on the context provided.
+    prompt = f"""You are a highly clever question-answering assistant trained to provide clear and concise answers based on the user query and provided context. 
+    Your task is to generate answers for the user query based on the context provided.
     Instructions for your response:
     1. Directly answer the user query. Make use of provided context if necessary.
     2. Ensure your response is clear and concise.
-    3. Give the answer in json format with a single key named 'content' with value as your response.
-    4. It is important to give response in json format, otherwise the chat may terminate.
-
+    3. Give the answer in JSON format with a single key named 'content' with value as your response.
+    4. It is important to give response in JSON format, otherwise the chat may terminate.
     Please format an answer based on the following user question and context provided:
-
     User Question: 
     {question}
-
     Context:
     {context}
     """
@@ -565,14 +377,27 @@ def reguar_answer_node(question, context):
     res = response.content
     return res
 
+def general_answer_node(question):
+    prompt = f"""You are an assistant that provides helpful and accurate answers to any question. Please answer the following question in a JSON format with a single key 'content' containing your answer.
+
+    Question: {question}
+    """
+    response = llm.invoke(
+        [SystemMessage(content=prompt)]
+    )
+    try:
+        res = json.loads(response.content)
+        return res
+    except json.JSONDecodeError:
+        # If the response is not valid JSON, wrap it
+        return {'content': response.content}
+
 CURR_CONTEXT = ''
 
-# @spaces.GPU
 def get_answer(image=[], message='', sessionID='abc123'):
     global CURR_CONTEXT
     if len(image) > 0:
         try:
-            # Process the image and message here
             device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
             chat = Chat(model,transform,df,tar_img_feats, device)
             chat.encode_image(image)
@@ -587,29 +412,34 @@ def get_answer(image=[], message='', sessionID='abc123'):
             print(e)
             response = {'content':"An error occurred while processing your request."}
     elif len(image) == 0 and message is not None:
-        print("I am here")
         task = router_node(message)
         if task == 'retrieval':
+            formated_input = {
+                'input': message,
+                'context': CURR_CONTEXT
+            }
+            response = json_answer_generator(message, CURR_CONTEXT)
+        elif task == "recommendation":
             recipes = recommendation_node(message)
-            print(recipes)
             if not recipes:
                 response = {'content':"An error occurred while processing your request."}
             response = answer_formatter_node(message, recipes)
+        elif task == "general":
+            response = general_answer_node(message)
+            if response is None:
+                response = {'content':"An error occurred while processing your request."}
         else:
-            formated_input = {
-                'input': message,
-                'context': CURR_CONTEXT
-            }
-            response = json_answer_generator(message, data)
+            response = {'content':"Sorry, I didn't understand your request."}
+    else:
+        response = {'content':"Please provide a message to process."}
 
     return response
 
 # Function to handle WebSocket connection
 @socketio.on('ping')
-def handle_connect():
+def handle_ping():
     emit('Ping-return', {'message': 'Connected'}, room=request.sid)
 
-
 # Function to handle WebSocket connection
 @socketio.on('connect')
 def handle_connect():
@@ -620,9 +450,7 @@ def handle_connect():
 def handle_disconnect():
     print(f"Client disconnected: {request.sid}")
 
-import json
 import base64
-from PIL import Image
 from io import BytesIO
 import torchvision.transforms as transforms
 
@@ -655,7 +483,6 @@ def handle_message(data):
         if session_store[session_id]['image_data'] or session_store[session_id]['message']:
             try:
                 image_bytes = session_store[session_id]['image_data']
-                # print("checkpoint 2")
                 if isinstance(image_bytes, str):
                     image_bytes = base64.b64decode(image_bytes)
                 image = Image.open(BytesIO(image_bytes))
@@ -670,7 +497,6 @@ def handle_message(data):
                     'input': message,
                     'context': json.dumps(context)
                 }
-                # Invoke question_answer_chain and stream the response
                 response = json_answer_generator(message, context)
                 emit('response', response, room=session_id)
 
@@ -687,74 +513,25 @@ def handle_message(data):
         if task == 'retrieval':
             formated_input = {
                 'input': message,
-                'context': json.dumps(CURR_CONTEXT)
+                'context': CURR_CONTEXT
             }
             response = json_answer_generator(message, CURR_CONTEXT)
             emit('response', response, room=session_id)
         elif task == "recommendation":
-            response = recommendation_node(message)
-            # response = answer_formatter_node(message, recipes)
-            if response is None:
+            recipes = recommendation_node(message)
+            if not recipes:
                 response = {'content':"An error occurred while processing your request."}
-
+            response = answer_formatter_node(message, recipes)
             emit('json_response', response, room=session_id)
-        else:
-            response = reguar_answer_node(message, CURR_CONTEXT)
+        elif task == "general":
+            response = general_answer_node(message)
             if response is None:
                 response = {'content':"An error occurred while processing your request."}
             emit('json_response', response, room=session_id)
+        else:
+            response = {'content':"Sorry, I didn't understand your request."}
+            emit('json_response', response, room=session_id)
         session_store.pop(session_id, None)
-    
-
-
-import requests
-from PIL import Image
-import numpy as np
-from io import BytesIO
-
-def download_image_to_numpy(url):
-    print("Image URL: ", url)
-    # Send a GET request to the URL to download the image
-    response = requests.get(url)
-    
-    # Check if the request was successful
-    if response.status_code == 200:
-        # Open the image using PIL and convert it to RGB format
-        image = Image.open(BytesIO(response.content)).convert('RGB')
-        
-        # Convert the image to a NumPy array
-        image_array = np.array(image)
-        
-        return image_array
-    else:
-        raise Exception(f"Failed to download image. Status code: {response.status_code}")
-
-import base64
-import numpy as np
-from io import BytesIO
-from PIL import Image
-
-def base64_to_numpy(base64_string):
-    # Decode the base64 string
-    image_data = base64.b64decode(base64_string)
-    
-    # Convert the byte data to a PIL image
-    image = Image.open(BytesIO(image_data))
-    
-    # Convert the PIL image to a NumPy array
-    image_np = np.array(image)
-    
-    return image_np
-
-@socketio.on('example')
-def handle_message(data):
-    img_url = data['img_url']
-    message = data['message']
-    session_id = request.sid
-    image_array = base64_to_numpy(img_url)
-    response = get_answer(image=image_array, message=message, sessionID=request.sid)
-    emit('response', response, room=session_id)
-    return response
 
 # Home route
 @app.route("/")
@@ -763,4 +540,4 @@ def index_view():
 
 # Main function to run the app
 if __name__ == '__main__': 
-    socketio.run(app, debug=True)
+    socketio.run(app, debug=True)