Update app.py

app.py

@@ -374,13 +374,15 @@ def determine_tool_usage(query: str) -> str:
 
 
 # Function to generate text using the LLM, incorporating tool results if available
-def generate_text(prompt: str, tool_results: dict = None) -> str:
+def generate_text(prompt: str, tool_results: dict = None, chat_history: list[dict] = None) -> str:
     """
-    Generates text using the configured LLM, optionally incorporating tool results.
+    Generates text using the configured LLM, optionally incorporating tool results and chat history.
     Args:
-        prompt: The initial prompt for the LLM.
+        prompt: The initial prompt for the LLM (the user's latest query).
         tool_results: A dictionary containing results from executed tools.
                       Keys are tool names, values are their outputs.
+        chat_history: The history of the conversation as a list of dictionaries
+                      (as provided by Gradio ChatInterface with type="messages").
     Returns:
         The generated text from the LLM.
     """
@@ -389,34 +391,52 @@ def generate_text(prompt: str, tool_results: dict = None) -> str:
 You are a friendly and helpful chatbot. Respond to greetings appropriately (e.g., "Hello!", "Hi there!", "Habari!"). If the user uses Swahili greetings or simple conversational phrases, respond in Swahili. Otherwise, respond in English unless the query is clearly in Swahili. Handle conversational flow and ask follow-up questions when appropriate.
 If the user asks a question about other companies or general knowledge, answer their question. However, subtly remind them that your primary expertise and purpose are related to Absa-specific information.
 """
-    full_prompt_builder = [persona_instructions, prompt]
+    # Build the messages list for the chat completion API
+    messages = [{"role": "user", "content": persona_instructions}] # Start with the persona instructions
 
+    if chat_history:
+        print("Including chat history in LLM prompt.")
+        # Iterate through the chat_history provided by Gradio (list of dictionaries)
+        # Add only 'user' and 'assistant' roles to the LLM context
+        for message_dict in chat_history:
+            role = message_dict.get("role")
+            content = message_dict.get("content")
+            if role in ["user", "assistant"] and content is not None:
+                messages.append({"role": role, "content": content})
+
+
+    # Add the current user prompt and tool results
+    current_user_content = prompt
     if tool_results and any(tool_results.values()):
-        full_prompt_builder.append("\n\nTool Results:\n")
-        for question, results in tool_results.items(): # Iterate through results per question
+        current_user_content += "\n\nTool Results:\n"
+        for question, results in tool_results.items():
             if results:
-                full_prompt_builder.append(f"--- Results for: {question} ---\n") # Add question context
+                current_user_content += f"--- Results for: {question} ---\n"
                 if isinstance(results, list):
                     for i, result in enumerate(results):
-                         # Check if the result is from business info retrieval
-                         if isinstance(result, dict) and 'Service' in result and 'Description' in result:
-                             full_prompt_builder.append(f"Business Info {i+1}:\nService: {result.get('Service', 'N/A')}\nDescription: {result.get('Description', 'N/A')}\n\n")
-                         elif isinstance(result, dict) and 'url' in result: # Check if the result is from DuckDuckGo
-                             full_prompt_builder.append(f"Search Result {i+1}:\nTitle: {result.get('title', 'N/A')}\nURL: {result.get('url', 'N/A')}\nSnippet: {result.get('body', 'N/A')}\n\n")
-                         else:
-                              full_prompt_builder.append(f"{result}\n\n") # Handle other list items
+                        if isinstance(result, dict) and 'Service' in result and 'Description' in result:
+                            current_user_content += f"Business Info {i+1}:\nService: {result.get('Service', 'N/A')}\nDescription: {result.get('Description', 'N/A')}\n\n"
+                        elif isinstance(result, dict) and 'url' in result:
+                            current_user_content += f"Search Result {i+1}:\nTitle: {result.get('title', 'N/A')}\nURL: {result.get('url', 'N/A')}\nSnippet: {result.get('body', 'N/A')}\n\n"
+                        else:
+                            current_user_content += f"{result}\n\n"
                 elif isinstance(results, dict):
-                     for key, value in results.items():
-                         full_prompt_builder.append(f"{key}: {value}\n")
-                     full_prompt_builder.append("\n")
+                    for key, value in results.items():
+                        current_user_content += f"{key}: {value}\n"
+                    current_user_content += "\n"
                 else:
-                     full_prompt_builder.append(f"{results}\n\n") # Handle single string results (like date calculation)
-        full_prompt_builder.append("Based on the provided tool results, answer the user's original query. If a question was answered by a tool, use the tool's result directly in your response. Maintain the language of the original query if possible, especially for simple greetings or direct questions answered by tools.")
+                    current_user_content += f"{results}\n\n"
+
+        current_user_content += "Based on the provided tool results and the conversation history, answer the user's latest query. If a question was answered by a tool, use the tool's result directly in your response. Maintain the language of the original query if possible, especially for simple greetings or direct questions answered by tools."
         print("Added tool results and instruction to final prompt.")
     else:
-        print("No tool results to add to final prompt.")
-    full_prompt = "".join(full_prompt_builder)
-    print(f"Sending prompt to LLM:\n---\n{full_prompt}\n---")
+         current_user_content += "Based on the conversation history, answer the user's latest query."
+         print("No tool results to add to final prompt, relying on conversation history.")
+
+    messages.append({"role": "user", "content": current_user_content})
+
+
+    print(f"Sending messages to LLM:\n---\n{messages}\n---")
     generation_config = {
         "temperature": 0.7,
         "max_new_tokens": 500,
@@ -426,9 +446,7 @@ If the user asks a question about other companies or general knowledge, answer t
     }
     try:
         response = client.chat_completion(
-            messages=[
-                {"role": "user", "content": full_prompt}
-            ],
+            messages=messages, # Pass the list of messages
             max_tokens=generation_config.get("max_new_tokens", 512),
             temperature=generation_config.get("temperature", 0.7),
             top_p=generation_config.get("top_p", 0.95)
@@ -517,8 +535,8 @@ Query: {query}
 
     # Step 5: Final Response Generation
     print("\n--- Generating final response ---")
-    # The generate_text function already handles incorporating tool results if provided
-    final_response = generate_text(query, tool_results)
+    # Pass the chat_history (which is a list of dictionaries when using type="messages")
+    final_response = generate_text(query, tool_results, chat_history)
     print("\n--- Final Response from LLM ---")
     print(final_response)
     print("\n----------------------------")