do it

app/main.py

@@ -146,10 +146,10 @@ def get_llm(model_name: str):
 ###############################################################################
 
 @app.get("/suggestions", response_model=List[Dict[str, Any]])
-async def get_suggestions() -> List[Dict[str, Any]]:
+async def get_suggestions(model: str = "Mistral") -> List[Dict[str, Any]]:
     """
     For each fixed business question, generate three concise bullet-point suggestions.
-    This endpoint uses gpt-4o-mini-2024-07-18 via LangChain.
+    This endpoint can use various models via Hugging Face Inference API.
     """
     suggestions_list = []
     
@@ -162,11 +162,9 @@ async def get_suggestions() -> List[Dict[str, Any]]:
         template=suggestion_template_str
     )
     
-    # Use ChatOpenAI (gpt-4o-mini-2024-07-18) to generate suggestions.
-    suggestion_chain = LLMChain(
-        llm=ChatOpenAI(model_name="gpt-4o-mini-2024-07-18", temperature=0.7), 
-        prompt=suggestion_prompt
-    )
+    # Get the appropriate LLM based on the selected model
+    llm = get_llm(model)
+    suggestion_chain = LLMChain(llm=llm, prompt=suggestion_prompt)
     
     for question in QUESTIONS:
         try: