fix: Remove internal prompt instructions from user-facing responses

- Updated prompt constructions in `handle_query` and `AnswerExpander` to use direct instructions without conversational preambles.
- Modified `GeminiLLM` class to utilize system and user roles, ensuring internal instructions are not echoed back to users.
- Ensured that all responses returned to users are free from embedded prompt instructions.
- Enhanced prompt clarity for better content generation and user experience.
- Maintained existing functionalities such as caching, retrieval, and web search integrations.

app.py

@@ -78,8 +78,20 @@ class GeminiLLM(LLM):
             chat_session = model.start_chat(history=[])
             logger.debug("Chat session started.")
 
-            response = chat_session.send_message(prompt)
-            logger.debug(f"Prompt sent to model: {prompt}")
+            # Use role-based messages if supported
+            system_message = {
+                "role": "system",
+                "content": "You are Daily Wellness AI, a friendly and professional wellness assistant."
+            }
+            user_message = {
+                "role": "user",
+                "content": prompt
+            }
+
+            chat_session.send_message(system_message)
+            chat_session.send_message(user_message)
+
+            response = chat_session.get_response()
             logger.debug(f"Raw response received: {response.text}")
 
             return response.text
@@ -187,7 +199,7 @@ class QuestionSanityChecker:
 
     def is_relevant(self, question: str) -> bool:
         prompt = (
-            f"You are an assistant that determines whether a question is relevant to daily wellness.\n\n"
+            f"Determine whether the following question is relevant to daily wellness.\n\n"
             f"Question: {question}\n\n"
             f"Is the above question relevant to daily wellness? Respond with 'Yes' or 'No' only."
         )
@@ -255,18 +267,16 @@ class AnswerExpander:
                 "Provide a thorough, in-depth explanation, adding relevant tips and context, "
                 "while remaining creative and brand-aligned. "  
                 if detail else
-                "Please provide a concise response in no more than 4 sentences."
+                "Provide a concise response in no more than 4 sentences."
             )
 
             prompt = (
-                f"You are Daily Wellness AI, a friendly wellness expert. Below are multiple "
-                f"potential answers retrieved from a local knowledge base. You have a user question.\n\n"
+                f"Synthesize the following retrieved answers into a single cohesive, creative, and brand-aligned response. "
+                f"{detail_instructions} "
+                f"Conclude with a short inspirational note.\n\n"
                 f"Question: {query}\n\n"
                 f"Retrieved Answers:\n{reference_block}\n\n"
-                f"Please synthesize these references into a single cohesive, creative, and brand-aligned response. "
-                f"{detail_instructions} "
-                f"End with a short inspirational note.\n\n"
-                "Disclaimer: This is general wellness information, not a substitute for professional medical advice."
+                "Disclaimer: This is general wellness information and not a substitute for professional medical advice."
             )
 
             logger.debug(f"Generated prompt for answer expansion: {prompt}")
@@ -402,12 +412,10 @@ def handle_query(query: str, detail: bool = False) -> str:
             # Combine any cached answer (if it exists) with the web result
             if cached_answer:
                 blend_prompt = (
-                    f"You previously provided the following answer:\n\n"
-                    f"{cached_answer}\n\n"
-                    f"Now, we also have these web results:\n\n"
-                    f"{web_search_response}\n\n"
-                    "Please combine them into a more creative and accurate response. "
-                    "Add positivity and end with a short inspirational note."
+                    f"Combine the following previous answer with the new web results to create a more creative and accurate response. "
+                    f"Add positivity and conclude with a short inspirational note.\n\n"
+                    f"Previous Answer:\n{cached_answer}\n\n"
+                    f"Web Results:\n{web_search_response}"
                 )
                 final_answer = llm._call(blend_prompt).strip()
             else:
@@ -424,16 +432,15 @@ def handle_query(query: str, detail: bool = False) -> str:
             return final_answer
 
         # 6) If similarity is sufficient, we will finalize an answer from the knowledge base
-        responses = [ans[0] for ans in retrieved]
+        responses = [ans for ans, score in retrieved]
 
         # 6a) If we have a cached answer, let's blend it with the new knowledge base data
         if cached_answer:
             blend_prompt = (
-                f"You provided this answer previously:\n\n"
-                f"{cached_answer}\n\n"
-                f"Now, here are some newly retrieved answers:\n\n"
-                f"{chr(10).join(f'- {r}' for r in responses)}\n\n"
-                "Please synthesize these together, adding new insights, creativity, and a short inspirational note at the end."
+                f"Combine the previous answer with the newly retrieved answers to enhance creativity and accuracy. "
+                f"Add new insights, creativity, and conclude with a short inspirational note.\n\n"
+                f"Previous Answer:\n{cached_answer}\n\n"
+                f"New Retrieved Answers:\n" + "\n".join(f"- {r}" for r in responses)
             )
             final_answer = llm._call(blend_prompt).strip()
         else:
@@ -449,7 +456,6 @@ def handle_query(query: str, detail: bool = False) -> str:
         logger.debug("Exception details:", exc_info=True)
         return "An error occurred while processing your request."
 
-
 ###############################################################################
 # 11) Gradio Interface
 ###############################################################################