Sanity check

app.py

@@ -173,7 +173,33 @@ class EmbeddingRetriever:
 retriever = EmbeddingRetriever(corpus_questions, corpus_answers, question_embeddings, embedding_model, cross_encoder)
 
 ###############################################################################
-# 6) Answer Expansion
+# 6) Sanity Check Tool
+###############################################################################
+class QuestionSanityChecker:
+    def __init__(self, llm: GeminiLLM):
+        self.llm = llm
+
+    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"Question: {question}\n\n"
+            f"Is the above question relevant to daily wellness? Respond with 'Yes' or 'No' only."
+        )
+        try:
+            response = self.llm._call(prompt)
+            is_yes = 'yes' in response.lower()
+            logger.debug(f"Sanity check response: {response}, interpreted as {is_yes}")
+            return is_yes
+        except Exception as e:
+            logger.error(f"Error in sanity check: {e}")
+            logger.debug("Exception details:", exc_info=True)
+            return False
+
+# Instantiate the sanity checker globally
+sanity_checker = QuestionSanityChecker(llm)
+
+###############################################################################
+# 7) Answer Expansion
 ###############################################################################
 class AnswerExpander:
     def __init__(self, llm: GeminiLLM):
@@ -183,14 +209,15 @@ class AnswerExpander:
         try:
             reference_block = "\n".join(f"- {idx+1}) {ans}" for idx, ans in enumerate(retrieved_answers, start=1))
             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"Question: {query}\n\n"
-              f"Retrieved Answers:\n{reference_block}\n\n"
-              "Please synthesize these references into a single cohesive, creative, and brand-aligned response. "
-              "Add practical tips and positivity, and end with a short inspirational note. "
-              "Please provide a concise response in no more than 4 sentences.\n\n"
-              "Disclaimer: This is general wellness information, not a substitute for professional medical advice.")
+                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"Question: {query}\n\n"
+                f"Retrieved Answers:\n{reference_block}\n\n"
+                "Please synthesize these references into a single cohesive, creative, and brand-aligned response. "
+                "Add practical tips and positivity, and end with a short inspirational note. "
+                "Please provide a concise response in no more than 4 sentences.\n\n"
+                "Disclaimer: This is general wellness information, not a substitute for professional medical advice."
+            )
             logger.debug(f"Generated prompt for answer expansion: {prompt}")
             response = self.llm._call(prompt)
             logger.debug(f"Expanded answer: {response}")
@@ -203,16 +230,31 @@ class AnswerExpander:
 answer_expander = AnswerExpander(llm)
 
 ###############################################################################
-# 7) Query Handling
+# 8) Query Handling
 ###############################################################################
 def handle_query(query: str) -> str:
     if not query or not isinstance(query, str) or len(query.strip()) == 0:
         return "Please provide a valid question."
 
     try:
+        # Sanity Check: Determine if the question is relevant to daily wellness
+        is_relevant = sanity_checker.is_relevant(query)
+        if not is_relevant:
+            return "Your question seems out of context or not related to daily wellness. Please ask a wellness-related question."
+
+        # Proceed with retrieval
         retrieved = retriever.retrieve(query)
         if not retrieved:
             return "I'm sorry, I couldn't find an answer to your question."
+
+        # Optional: Check similarity threshold (if still desired)
+        top_score = retrieved[0][1]  # Assuming the list is sorted descending
+        similarity_threshold = 0.3  # Adjust this threshold based on empirical results
+
+        if top_score < similarity_threshold:
+            return "I'm sorry, I didn't understand your question. Could you please rephrase it?"
+
+        # Proceed with answer expansion
         responses = [ans[0] for ans in retrieved]
         expanded_answer = answer_expander.expand(query, responses)
         return expanded_answer
@@ -222,7 +264,7 @@ def handle_query(query: str) -> str:
         return "An error occurred while processing your request."
 
 ###############################################################################
-# 8) Gradio Interface
+# 9) Gradio Interface
 ###############################################################################
 def gradio_interface(query: str):
     try:
@@ -260,7 +302,7 @@ interface = gr.Interface(
 )
 
 ###############################################################################
-# 9) Launch Gradio
+# 10) Launch Gradio
 ###############################################################################
 if __name__ == "__main__":
     try: