integrated image description'

api/routers/analytics_chatbot.py

@@ -2,15 +2,19 @@ import ast
 import os
 import json
 import hashlib
+import base64
 import logging
 from fastapi import APIRouter
+from typing import Optional
 from redis import Redis
 from fastapi.responses import StreamingResponse
 from api.stored_data import stored_data
 from src.genai.analytics_chatbot.agent import ChatbotAgent
 from src.genai.analytics_chatbot.utils.utils import process_query
-from api.schemas.analytics_chatbot import UserMessage
+from src.genai.analytics_chatbot.handlers.analytics_description import generate_analytics_description
+from api.schemas.analytics_chatbot import AnalyticsChatRequest
 from config.redis_config import redis_client
+from openai import OpenAI
 
 
 app_logger = logging.getLogger("app_logger")
@@ -21,29 +25,38 @@ router = APIRouter()
 agent=ChatbotAgent()
 graph = agent.chatbot_graph() 
 
-@router.get("/analytics-chatbot")
-def get_analytics(msg:str):
-    # user_query=process_query(msg)
-    cache_key = f"analytics:{hashlib.md5(msg.encode()).hexdigest()}"
+
+@router.post("/analytics-chatbot")
+def get_analytics(request: AnalyticsChatRequest):
+    user_query = process_query(request.msg)
+    print('Processed user query:', user_query)
     
+    cache_key = f"analytics:{hashlib.md5(user_query.encode()).hexdigest()}"
     cached_response = redis_client.get(cache_key)
-    if cached_response:
-        print("Cache hit")
-        return json.loads(cached_response)
-    
-    config={"configurable": {"thread_id": "analytics-chatbot-thread"},"run_name":"analytics-chatbot"}
-    result=graph.invoke({'messages':msg},config=config)
+    print('cached-response:', cached_response)
 
-    print('Final Result:', result)
-    response_to_cache = {}
-    if result.get('backup_data') is not None:
-        response_to_cache = {'backup_response': result['backup_data']}
+    if cached_response:
+        response_to_cache = json.loads(cached_response)
     else:
-        response_to_cache = {
-            'response': result['response'],
-            'endpoint': result['endpoint']
-        }
+        response_to_cache = {}
+
+    if not response_to_cache.get('response') or not response_to_cache.get('endpoint'):
+        config = {"configurable": {"thread_id": "analytics-chatbot-thread"},
+                  "run_name": "analytics-chatbot"}
+        result = graph.invoke({'messages': user_query}, config=config)
 
-    # Save to Redis with TTL (e.g., 1 hour)
-    redis_client.set(cache_key, json.dumps(response_to_cache), ex=1)  # ex=seconds
+        if result.get('backup_data') is not None:
+            response_to_cache['backup_response'] = result['backup_data']
+        else:
+            response_to_cache['response'] = result['response']
+            response_to_cache['endpoint'] = result['endpoint']
+
+    if request.image_base64 and not response_to_cache.get('description'):
+        description = generate_analytics_description(user_query, request.image_base64)
+        response_to_cache['description'] = description
+
+    redis_client.set(cache_key, json.dumps(response_to_cache), ex=3000)
     return response_to_cache
+
+    
+

api/schemas/analytics_chatbot.py

@@ -1,4 +1,6 @@
 from pydantic import BaseModel
+from typing import Optional
 
-class UserMessage(BaseModel):
-    message: str
+class AnalyticsChatRequest(BaseModel):
+    msg: str
+    image_base64:Optional[str] = None

src/genai/analytics_chatbot/handlers/__init__.py

@@ -0,0 +1,15 @@
+from .compare import compare
+from .posting_time import get_posting_time
+from .peak_comment_hour import get_peak_comment_hour
+from .emoji_count import get_emoji_count
+from .comment_quality import get_comment_quality
+from .bot_and_diversity import get_bot_and_diversity
+
+__all__ = [
+    "compare",
+    "get_posting_time",
+    "get_peak_comment_hour",
+    "get_emoji_count",
+    "get_comment_quality",
+    "get_bot_and_diversity",
+]

src/genai/analytics_chatbot/handlers/analytics_description.py

@@ -0,0 +1,31 @@
+
+from openai import OpenAI
+import os
+from ..utils.prompts import analytics_description_prompt
+client = OpenAI(
+base_url="https://router.huggingface.co/v1",
+api_key=os.environ['HUGGINGFACEHUB_ACCESS_TOKEN'],)
+
+def generate_analytics_description(query,image_base64):
+    completion = client.chat.completions.create(
+        model="zai-org/GLM-4.5V:novita",
+        messages=[
+            {
+                "role": "user",
+                "content": [
+                    {
+                        "type": "text",
+                        "text": analytics_description_prompt(query)
+                    },
+                    {
+                        "type": "image_url",
+                        "image_url": {
+                            "url": f"data:image/jpeg;base64,{image_base64}"
+                        }
+                    }
+                ]
+            }
+        ],
+    )
+
+    return completion.choices[0].message.content

src/genai/analytics_chatbot/handlers/comment_quality.py

@@ -13,5 +13,6 @@ def get_comment_quality(state,llm_gpt,url):
     parameters=llm_gpt.with_structured_output(CommentQualityFormatter , method='function_calling').invoke(messages)
     print(parameters)
     response = requests.get(url, params={ 'start_date': parameters.start_date , 'end_date':parameters.end_date,'influencer_username':process_query(parameters.influencer_name)})
+    print(response)
     return response
 

src/genai/analytics_chatbot/utils/name_variations.json

@@ -1155,5 +1155,11 @@
     "tashyil ha g",
     "tashyil",
     "tashyil g"
-  ]
+  ],
+  "good sentiment":[
+    "good"
+  ],
+  "bad sentiment": [
+    "bad"
+]
 }

src/genai/analytics_chatbot/utils/nodes.py

@@ -9,12 +9,14 @@ from src.genai.utils.models_loader import llm_gpt
 import numpy as np
 from src.genai.utils.data_loader import api_knowledge_df, api_index, caption_df , caption_index
 from src.genai.utils.models_loader import embedding_model
-from ..handlers.compare import compare
-from ..handlers.posting_time import get_posting_time
-from ..handlers.peak_comment_hour import get_peak_comment_hour
-from ..handlers.emoji_count import get_emoji_count
-from ..handlers.comment_quality import get_comment_quality
-
+from ..handlers import (
+    compare,
+    get_posting_time,
+    get_peak_comment_hour,
+    get_emoji_count,
+    get_comment_quality,
+    get_bot_and_diversity,
+)
 
 
 class FetchLastMessage:
@@ -39,12 +41,11 @@ class RetrievePossibleEndpoints:
         self.df = api_knowledge_df
         self.index = api_index
         self.results = []
-        # self.results = ['/api/v1/compare/', '/api/v1/engagement/basic-metrics', '/api/v1/content/hashtags-analysis', '/api/v1/audience/emoji-count', '/api/v1/engagement/temporal_analysis']
 
     def run(self,state:State):
         print('Gone to retrieve possible endpoints')
         query_embedding = np.array(embedding_model.embed_query(state['latest_message'])).reshape(1, -1).astype('float32')
-        distances, indices = self.index.search(query_embedding, 5)
+        distances, indices = self.index.search(query_embedding,10)
         for idx in indices[0]:
             row = self.df.iloc[idx]
             print('Endpoint:',row['endpoint'])
@@ -135,6 +136,15 @@ class FetchDataNode:
             "Authorization": "Bearer YOUR_API_KEY",  # replace with your API key if needed
             "Content-Type": "application/json"
             }
+        
+        self.endpoint_handlers = {
+            '/api/v1/compare/': compare,
+            '/api/v1/engagement/posting-time-analysis': get_posting_time,
+            '/api/v1/audience/peak-comment-hour': get_peak_comment_hour,
+            '/api/v1/audience/emoji-count': get_emoji_count,
+            '/api/v1/audience/comment-quality': get_comment_quality,
+            '/api/v1/audience/bot-and-diversity': get_bot_and_diversity
+        }
 
     def run(self, state:State):
         try:
@@ -142,26 +152,14 @@ class FetchDataNode:
             print('Entered to fetch data')
             url = f'''{self.base_url}{state['endpoint']}'''
 
-            if state['endpoint'] == '/api/v1/compare/':
-                response=compare(state,llm_gpt,url)
-                return {'response': response.json()}
-            
-            elif state['endpoint'] == '/api/v1/engagement/posting-time-analysis':
-                response = get_posting_time(state, llm_gpt,url)
-                return {'response': response.json()}
-            
-            elif state['endpoint']=='/api/v1/audience/peak-comment-hour':
-                response = get_peak_comment_hour(state,llm_gpt,url)
-                return {'response':response.json()}
-            
-            elif state['endpoint']== '/api/v1/audience/emoji-count':
-                response = get_emoji_count(state,llm_gpt,url)
-                return {'response':response.json()}
-            
-            elif state['endpoint']== '/api/v1/audience/comment-quality':
-                response = get_comment_quality(state,llm_gpt,url)
+            if state['endpoint'] in self.endpoint_handlers:
+                print('Entered to handler.')
+                handler = self.endpoint_handlers[state['endpoint']]
+                response = handler(state, llm_gpt, url)
+                print('Returned by handler.')
                 return {'response':response.json()}
 
+
             elif 'single_influencer_query' in  state['query_type']:
                 response = requests.get(url, params=state['parameters_values'],headers=self.headers)
                 print('Data from api:', response)

src/genai/analytics_chatbot/utils/prompts.py

@@ -25,7 +25,9 @@ output:
 '''
 
 fetch_last_message_prompt = '''
-You are an AI assistant that reads an entire conversation between a human and an AI. Your task is to detect the human's most recent intention, taking into account the full conversation history.  
+You are an AI assistant that reads an entire conversation between a human and an AI. 
+The human is trying to ask something about the influencers.
+Your task is to detect the human's most recent intention, taking into account the full conversation history.  
 
 - Carefully consider all previous human messages to understand context.  
 - Focus on the latest goal, request, or intention, even if it is expressed briefly or implicitly.  
@@ -108,4 +110,11 @@ You are perfect parameters extractor for analysis of bot and comment diversity o
 Given a user query and a list of needed parameters, return a Python dictionary assigning the best value for each parameter. 
 You have to return a dictionary containing influencer_name , number of commentors (top_n), start_date and  end_date from the user query.
 If there is no any specific mention of dates, you can return None for dates. In the case of number of commentors, return a default value of 10 if the number is not passed from the user.
+'''
+
+def analytics_description_prompt(query):
+    return f'''
+You are provided with the user query and the image. 
+Give clear information to the user for their query only by analyzing the image. Don't give any responses outside of the image's context.
+The user query is: \n{query}\n
 '''