Refactor import statements to use absolute paths and add a new prompt template for handling rude queries

py-src/lets_talk/agent.py

@@ -50,7 +50,7 @@ def call_model(model, state: Dict[str, Any]) -> Dict[str, list[BaseMessage]]:
         context = state.get("context", "")
         
         # Insert system message with context before the latest user message
-        from prompts import call_llm_prompt_template
+        from lets_talk.prompts import call_llm_prompt_template
         sys_prompt = call_llm_prompt_template.format(
             context=context,
         )
@@ -79,47 +79,47 @@ def call_model(model, state: Dict[str, Any]) -> Dict[str, list[BaseMessage]]:
         return {"messages": [HumanMessage(content=error_msg)]}
 
 
-def call_model(model, state: Dict[str, Any]) -> Dict[str, list[BaseMessage]]:
-    """
-    Process the current state through the language model.
+# def call_model(model, state: Dict[str, Any]) -> Dict[str, list[BaseMessage]]:
+#     """
+#     Process the current state through the language model.
     
-    Args:
-        model: Language model with tools bound
-        state: Current state containing messages and context
+#     Args:
+#         model: Language model with tools bound
+#         state: Current state containing messages and context
         
-    Returns:
-        Updated state with model's response added to messages
-    """
-    try:
-        messages = state["messages"]
-        context = state.get("context", "")
+#     Returns:
+#         Updated state with model's response added to messages
+#     """
+#     try:
+#         messages = state["messages"]
+#         context = state.get("context", "")
         
-        # Add context from documents if available
-        if context:
-            # Insert system message with context before the latest user message
-            context_message = SystemMessage(content=rag_prompt_template.format(context=context))
-
-            # Find the position of the last user message
-            for i in range(len(messages)-1, -1, -1):
-                if isinstance(messages[i], HumanMessage):
-                    # Insert context right after the last user message
-                    enhanced_messages = messages[:i+1] + [context_message] + messages[i+1:]
-                    break
-            else:
-                # No user message found, just append context
-                enhanced_messages = messages + [context_message]
-        else:
-            enhanced_messages = messages
+#         # Add context from documents if available
+#         if context:
+#             # Insert system message with context before the latest user message
+#             context_message = SystemMessage(content=rag_prompt_template.format(context=context))
+
+#             # Find the position of the last user message
+#             for i in range(len(messages)-1, -1, -1):
+#                 if isinstance(messages[i], HumanMessage):
+#                     # Insert context right after the last user message
+#                     enhanced_messages = messages[:i+1] + [context_message] + messages[i+1:]
+#                     break
+#             else:
+#                 # No user message found, just append context
+#                 enhanced_messages = messages + [context_message]
+#         else:
+#             enhanced_messages = messages
         
-        # Get response from the model
-        response = model.invoke(enhanced_messages)
-        return {"messages": [response]}
-    except Exception as e:
-        # Handle exceptions gracefully
-        error_msg = f"Error calling model: {str(e)}"
-        print(error_msg)  # Log the error
-        # Return a fallback response
-        return {"messages": [HumanMessage(content=error_msg)]}
+#         # Get response from the model
+#         response = model.invoke(enhanced_messages)
+#         return {"messages": [response]}
+#     except Exception as e:
+#         # Handle exceptions gracefully
+#         error_msg = f"Error calling model: {str(e)}"
+#         print(error_msg)  # Log the error
+#         # Return a fallback response
+#         return {"messages": [HumanMessage(content=error_msg)]}
 
 
 def should_continue(state: Dict[str, Any]) -> Union[Literal["action"], Literal["end"]]:

py-src/lets_talk/prompts.py

@@ -38,4 +38,49 @@ Query:
 # Output Format
 
 Respond only with "YES" or "NO".
-"""
+"""
+
+rude_query_answer_prompt_template = """\
+Respond to negative, rude, or derogatory questions or statements with respect, positivity, and an uplifting tone.
+
+Address the initial sentiment or statement with understanding and empathy before providing a positive response. Aim to uplift the conversation, converting any negative interaction into a positive engagement.
+
+# Steps
+
+1. Identify the negative or derogatory sentiment in the input.
+2. Acknowledge the sentiment or emotion behind the statement with empathy.
+3. Respond with positivity, focusing on encouraging and uplifting language.
+4. Conclude with a respectful and positive closing remark.
+
+# Output Format
+
+Respond using concise sentences or short paragraphs, maintaining a respectful and positive tone throughout.
+
+# Examples
+
+**Example 1:**
+
+- **Input:** "Go away"
+- **Output:** "I understand you might need some space, and I'm here to help whenever you're ready. Take care!"
+
+**Example 2:**
+
+- **Input:** "I am angry now"
+- **Output:** "It's okay to feel angry sometimes. If you need someone to talk to, I'm here for you, and we'll find a way through this together!"
+
+**Example 3:**
+
+- **Input:** "Tell me something emse"
+- **Output:** "Sure, I'd love to share something uplifting with you! Did you know that taking a moment to appreciate small things can brighten your day? :)"
+
+**Example 4:**
+
+- **Input:** "RIP you are awful"
+- **Output:** "I'm sorry if I disappointed you. I'm here to improve and assist you better. Let's turn this around together!"
+
+# Notes
+
+- Always maintain a positive and empathetic approach, even when the input is challenging.
+- Aim to uplift and provide encouragement, transforming the interaction into a positive experience.
+"""
+

py-src/lets_talk/rag.py

@@ -22,7 +22,7 @@ retriever = vector_store.as_retriever()
 llm = ChatOpenAI(model=config.LLM_MODEL, temperature=config.LLM_TEMPERATURE)
 
 
-from prompts import rag_prompt_template
+from lets_talk.prompts import rag_prompt_template
 
 rag_prompt = ChatPromptTemplate.from_template(rag_prompt_template)