Update app.py

Helper Function (call_llm): Introduced to handle responses from LiteLLMModel.complete(), ensuring that text is extracted correctly.
Sanity Checker & Answer Expander: Updated to use call_llm instead of direct model calls to avoid type errors.
Other Calls to LLM: Replaced direct calls like llm(prompt) within handle_query and other blending scenarios with call_llm(prompt).
General Adjustments: Ensured consistency in handling model responses throughout the code.

app.py

@@ -41,6 +41,18 @@ logger.info("GEMINI API Key loaded successfully.")
 # Instantiate the model using LiteLLMModel
 llm = LiteLLMModel(model_id="gemini/gemini-pro", api_key=gemini_api_key)
 
+###############################################################################
+# Helper Function for LLM Calls
+###############################################################################
+def call_llm(prompt: str) -> str:
+    """
+    Helper to call the LLM with a prompt, handling response extraction.
+    """
+    result = llm.complete(prompt)
+    if isinstance(result, dict):
+        return result.get('text', '')
+    return result
+
 ###############################################################################
 # 3) CSV Loading and Processing
 ###############################################################################
@@ -141,7 +153,7 @@ class QuestionSanityChecker:
             f"Is the above question relevant to daily wellness? Respond with 'Yes' or 'No' only."
         )
         try:
-            response = self.llm(prompt)
+            response = call_llm(prompt)
             is_yes = 'yes' in response.lower()
             is_no = 'no' in response.lower()
             logger.debug(f"Sanity check response: '{response}', interpreted as is_yes={is_yes}, is_no={is_no}")
@@ -162,25 +174,21 @@ sanity_checker = QuestionSanityChecker(llm)
 ###############################################################################
 # 7) smolagents Integration: GROQ Model and Web Search
 ###############################################################################
-# Initialize the smolagents' LiteLLMModel with GROQ model (already instantiated as llm if needed elsewhere)
-
 # Instantiate the DuckDuckGo search tool
 search_tool = DuckDuckGoSearchTool()
 
 # Create the web agent with the search tool
 web_agent = CodeAgent(
     tools=[search_tool],
-    model=llm  # Use the direct model for web queries if applicable
+    model=llm
 )
 
-# Define the managed web agent
 managed_web_agent = ManagedAgent(
     agent=web_agent,
     name="web_search",
     description="Runs a web search for you. Provide your query as an argument."
 )
 
-# Create the manager agent with managed web agent and additional tools if needed
 manager_agent = CodeAgent(
     tools=[],  # Add additional tools here if required
     model=llm,
@@ -219,7 +227,7 @@ class AnswerExpander:
             )
 
             logger.debug(f"Generated prompt for answer expansion: {prompt}")
-            response = self.llm(prompt)
+            response = call_llm(prompt)
             logger.debug(f"Expanded answer: {response}")
             return response.strip()
         except Exception as e:
@@ -328,7 +336,7 @@ def handle_query(query: str, detail: bool = False) -> str:
                     f"Previous Answer:\n{cached_answer}\n\n"
                     f"Web Results:\n{web_search_response}"
                 )
-                final_answer = llm(blend_prompt).strip()
+                final_answer = call_llm(blend_prompt).strip()
             else:
                 final_answer = (
                     f"**Daily Wellness AI**\n\n"
@@ -350,7 +358,7 @@ def handle_query(query: str, detail: bool = False) -> str:
                 f"Previous Answer:\n{cached_answer}\n\n"
                 f"New Retrieved Answers:\n" + "\n".join(f"- {r}" for r in responses)
             )
-            final_answer = llm(blend_prompt).strip()
+            final_answer = call_llm(blend_prompt).strip()
         else:
             final_answer = answer_expander.expand(query, responses, detail=detail)