Update app.py

app.py

@@ -18,22 +18,24 @@ from openai import OpenAI
 
 class BasicAgent:
     def __init__(self):
-        print("GAIA Smart Agent initialized")
+        print("Advanced GAIA Agent initialized")
         self.client = OpenAI()
 
+    # calculator tool
+    def calculate(self, expression):
+        try:
+            return str(eval(expression))
+        except:
+            return None
+
     def clean_answer(self, text):
-        """
-        GAIA scoring requires exact answers.
-        This removes explanations and keeps only the final value.
-        """
 
         text = text.strip()
 
-        # remove phrases
         text = re.sub(r"(?i)final answer[:\- ]*", "", text)
         text = re.sub(r"(?i)answer[:\- ]*", "", text)
 
-        # keep only last line
+        # take last line
         text = text.split("\n")[-1]
 
         return text.strip()
@@ -41,14 +43,13 @@ class BasicAgent:
     def ask_llm(self, question):
 
         prompt = f"""
-Solve the question carefully.
-
-Think step by step internally but output ONLY the final answer.
+You are solving a GAIA benchmark problem.
 
-Rules:
+Instructions:
+- Think step by step
+- Use math reasoning when necessary
+- Return ONLY the final answer
 - No explanation
-- No sentences
-- Only the final value
 
 Question:
 {question}
@@ -56,24 +57,46 @@ Question:
 
         response = self.client.chat.completions.create(
             model="gpt-4o-mini",
+            temperature=0,
             messages=[
-                {"role": "system", "content": "You are a precise problem solver."},
+                {"role": "system", "content": "You are a precise reasoning agent."},
                 {"role": "user", "content": prompt}
-            ],
-            temperature=0
+            ]
         )
 
         return response.choices[0].message.content.strip()
 
+    def try_math(self, question):
+
+        # detect simple math expressions
+        matches = re.findall(r"[\d\.\+\-\*\/\(\) ]+", question)
+
+        for m in matches:
+            m = m.strip()
+            if len(m) > 3:
+                result = self.calculate(m)
+                if result:
+                    return result
+
+        return None
+
     def __call__(self, question: str) -> str:
 
         print("Question:", question)
 
+        # try math tool first
+        math_result = self.try_math(question)
+
+        if math_result:
+            print("Calculator used:", math_result)
+            return math_result
+
+        # otherwise ask LLM
         raw_answer = self.ask_llm(question)
 
         final_answer = self.clean_answer(raw_answer)
 
-        print("Final Answer:", final_answer)
+        print("Final answer:", final_answer)
 
         return final_answer