Update app.py

app.py

@@ -50,7 +50,8 @@ class EnhancedContextDrivenChatbot:
         self.entity_tracker = {}
         self.conversation_context = ""
         self.model = model
-
+        self.last_instructions = None
+    
     def add_to_history(self, text):
         self.history.append(text)
         if len(self.history) > self.history_size:
@@ -79,6 +80,20 @@ class EnhancedContextDrivenChatbot:
         doc = nlp(text)
         return [chunk.text for chunk in doc.noun_chunks]
 
+    def extract_instructions(self, text):
+        instruction_patterns = [
+            r"^(?:please\s+)?(?:can\s+you\s+)?(?:could\s+you\s+)?(.*?)\s*(?:for\s+me|for\s+this\s+response|in\s+your\s+response|in\s+your\s+answer)(?:\s*\?)?$",
+            r"^(?:I\s+want\s+you\s+to\s+)?(.*?)\s*(?:for\s+me|for\s+this\s+response|in\s+your\s+response|in\s+your\s+answer)(?:\s*\?)?$",
+            r"^(?:make\s+sure\s+to\s+)?(.*?)\s*(?:for\s+me|for\s+this\s+response|in\s+your\s+response|in\s+your\s+answer)(?:\s*\?)?$",
+        ]
+        
+        for pattern in instruction_patterns:
+            match = re.match(pattern, text, re.IGNORECASE)
+            if match:
+                return match.group(1).strip(), True
+
+        return text, False
+
     def get_most_relevant_context(self, question):
         if not self.history:
             return question
@@ -100,16 +115,19 @@ class EnhancedContextDrivenChatbot:
         # Otherwise, it might be a new topic
         return question
 
-    def rephrase_query(self, question):
+    
+     def rephrase_query(self, question, instructions=None):
         if not self.model:
             return question  # Return original question if no model is available
         
+        instruction_prompt = f"Instructions: {instructions}\n" if instructions else ""
+        
         prompt = f"""
-        Given the conversation context and the current question, rephrase the question to include relevant context:
+        Given the conversation context, the current question, and any provided instructions, rephrase the question to include relevant context:
         
         Conversation context: {self.get_context()}
         Current question: {question}
-        
+        {instruction_prompt}
         Rephrased question:
         """
         
@@ -118,16 +136,21 @@ class EnhancedContextDrivenChatbot:
         return rephrased_question.strip()
 
     def process_question(self, question):
-        contextualized_question = self.get_most_relevant_context(question)
+        question, has_instructions = self.extract_instructions(question)
         
-        topics = self.extract_topics(question)
+        if has_instructions:
+            self.last_instructions = question
+            contextualized_question = self.get_most_relevant_context(question)
+        else:
+            contextualized_question = self.get_most_relevant_context(question)
+            if self.is_follow_up_question(question):
+                contextualized_question = self.rephrase_query(contextualized_question, self.last_instructions)
         
-        if self.is_follow_up_question(question):
-            contextualized_question = self.rephrase_query(contextualized_question)
+        topics = self.extract_topics(contextualized_question)
         
         self.add_to_history(question)
         
-        return contextualized_question, topics, self.entity_tracker
+        return contextualized_question, topics, self.entity_tracker, self.last_instructions
         
 # Initialize LlamaParse
 llama_parser = LlamaParse(
@@ -324,7 +347,7 @@ def ask_question(question, temperature, top_p, repetition_penalty, web_search, c
     context_reduction_factor = 0.7
 
     if web_search:
-        contextualized_question, topics, entity_tracker = chatbot.process_question(question)
+        contextualized_question, topics, entity_tracker, instructions = chatbot.process_question(question)
         serializable_entity_tracker = {k: list(v) for k, v in entity_tracker.items()}
         search_results = google_search(contextualized_question)
         all_answers = []
@@ -342,14 +365,17 @@ def ask_question(question, temperature, top_p, repetition_penalty, web_search, c
 
                 context_str = "\n".join([f"Source: {doc.metadata['source']}\nContent: {doc.page_content}" for doc in web_docs])
 
-                prompt_template = """
-                Answer the question based on the following web search results, conversation context, and entity information:
+                instruction_prompt = f"Instructions: {instructions}\n" if instructions else ""
+
+                prompt_template = f"""
+                Answer the question based on the following web search results, conversation context, entity information, and any provided instructions:
                 Web Search Results:
-                {context}
-                Conversation Context: {conv_context}
-                Current Question: {question}
-                Topics: {topics}
-                Entity Information: {entities}
+                {{context}}
+                Conversation Context: {{conv_context}}
+                Current Question: {{question}}
+                Topics: {{topics}}
+                Entity Information: {{entities}}
+                {instruction_prompt}
                 If the web search results don't contain relevant information, state that the information is not available in the search results.
                 Provide a summarized and direct answer to the question without mentioning the web search or these instructions.
                 Do not include any source information in your answer.