System prompt improving

app.py

@@ -81,7 +81,7 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
 
             try:
                 # Build new client with other provider and retry
-                agent = ReActAgent(provider="Google", model="gemini-2.5-pro")
+                agent = ReActAgent(provider="Mistral", model="mistral-large-latest")
                 submitted_answer = agent(str(item))
                 answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
                 results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
@@ -91,7 +91,7 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
                 print(f"Error running agent on task {task_id}: {e}")
 
                 try:
-                    agent = ReActAgent(provider="Mistral", model="mistral-large-latest")
+                    agent = ReActAgent(provider="Google", model="gemini-2.5-flash")
                     submitted_answer = agent(str(item))
                     answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
                     results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
@@ -127,8 +127,8 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
                                 print(f"Error running agent on task {task_id}: {e}")
                                 results_log.append({"Task ID": task_id, "Question": question_text, "Error": e})
     
-    # Wait for a while to ensure the agent is not overwhelmed
-    time.sleep(10)
+    # Wait for a while to ensure the model API doesn't reach the rate limit
+    time.sleep(60)
 
     if not answers_payload:
         print("Agent did not produce any answers to submit.")

react_agent.py

@@ -8,7 +8,7 @@ from langgraph.prebuilt import create_react_agent
 from custom_tools import custom_tools
 
 class ReActAgent:
-    def __init__(self, provider: str="Google", model: str="gemini-2.5-flash"):
+    def __init__(self, provider: str="Google", model: str="gemini-2.5-pro"):
         print('Initializing ReActAgent...')
         load_dotenv()
 
@@ -39,13 +39,24 @@ class ReActAgent:
                 max_retries=5
             )
 
-        sys_prompt = "You are a general AI assistant. I will ask you a question. Report your thoughts, and finish your answer with the following template: FINAL ANSWER: [YOUR FINAL ANSWER]. YOUR FINAL ANSWER should be a number OR as few words as possible OR a comma separated list of numbers and/or strings. If you are asked for a number, DON'T use comma to write your number NEITHER use units such as $ or percent sign unless specified otherwise. If you are asked for a string, DON'T use articles, NEITHER abbreviations (e.g. for cities) capitalize the first letter, and write the digits in plain text unless specified otherwise. If you are asked for a comma separated list, apply the above rules depending, unless the first letter capitalization, whether the element to be put in the list is a number or a string.\n\n\n \
-        \n \
-        You will be provided with tools to help you answer questions.\n \
-        If you are asked to make a calculation, absolutely use the tools provided to you. You should AVOID calculating by yourself and ABSOLUTELY use appropriate tools.\n \
-        If you need to search for information, use the web_search tool rather than wiki_search, unless the question specifies searching on wikipedia. After using the web_search tool, look for the first URL provided with the url_search tool and ask yourself if the answer is in the tool response. If it is, answer the question. If not, search on other links.\n \
-        \n \
-        If needed, use one tool first, then use the output of that tool as an input to another thinking then to the use of another tool."
+        sys_prompt = """
+        You are a general AI assistant. I will ask you a question. Report your thoughts, and finish your answer with the following template: FINAL ANSWER: [YOUR FINAL ANSWER].\n
+        \n
+        YOUR FINAL ANSWER should be a number OR as few words as possible OR a comma separated list of numbers and/or strings. If you are asked for a number, DON'T use comma to write your number NEITHER use units such as $ or percent sign unless specified otherwise. If you are asked for a string, DON'T use articles, NEITHER abbreviations (e.g. for cities) capitalize the first letter, and write the digits in plain text unless specified otherwise. If you are asked for a comma separated list, apply the above rules depending, unless the first letter capitalization, whether the element to be put in the list is a number or a string.\n
+        \n
+        EXAMPLES:\n
+        - What is US President Obama's first name? FINAL ANSWER: Barack\n
+        - What are the 3 mandatory ingredients for pancakes? FINAL ANSWER: eggs, flour, milk\n
+        - What is the final cost of an invoice comprising a $345.00 product and a $355.00 product? Provide the answer with two decimals. FINAL ANSWER: 700.00\n
+        - How many pairs of chromosomes does a human cell contain? FINAL ANSWER : 23\n
+        \n
+        \n
+        You will be provided with tools to help you answer questions.\n
+        If you are asked to make a calculation, absolutely use the tools provided to you. You should AVOID calculating by yourself and ABSOLUTELY use appropriate tools.\n
+        If you need to search for information, use the web_search tool rather than wiki_search, unless the question specifies searching on wikipedia. After using the web_search tool, look for the first URL provided with the url_search tool and ask yourself if the answer is in the tool response. If it is, answer the question. If not, search on other links.\n
+        \n
+        If needed, use one tool first, then use the output of that tool as an input to another thinking then to the use of another tool.
+        """
         # Build the ReAct agent
         self.agent = create_react_agent(
             model=llm,

requirements.txt

@@ -1,5 +1,4 @@
 beautifulsoup4==4.12.3
-duckduckgo-search==8.0.4
 gradio
 langchain==0.3.26 
 langchain_community==0.3.26