Update app.py

app.py

@@ -1,4 +1,5 @@
 import os
+import time
 import gradio as gr
 import requests
 import pandas as pd
@@ -23,7 +24,6 @@ def web_search(query: str) -> str:
     except Exception as e:
         return f"Search error: {e}"
 
-
 def wikipedia_search(query: str) -> str:
     try:
         search_url = "https://en.wikipedia.org/w/api.php"
@@ -48,7 +48,6 @@ def wikipedia_search(query: str) -> str:
     except Exception as e:
         return f"Wikipedia error: {e}"
 
-
 def run_python(code: str) -> str:
     import sys
     from io import StringIO
@@ -64,14 +63,13 @@ def run_python(code: str) -> str:
     finally:
         sys.stdout = old_stdout
 
-
 class SmartAgent:
     def __init__(self):
         self.api_key = os.getenv("GEMINI_API_KEY")
         if not self.api_key:
             raise ValueError("GEMINI_API_KEY environment variable not set!")
-        # Gemini 1.5 Flash - FREE tier: 1500 requests/day, 1M tokens/min
-        self.api_url = f"https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash-lite-preview-06-17:generateContent?key={self.api_key}"
+        # Valid Gemini 1.5 Flash endpoint string
+        self.api_url = f"https://generativelanguage.googleapis.com/v1beta/models/gemini-1.5-flash:generateContent?key={self.api_key}"
         print("SmartAgent initialized with Gemini 1.5 Flash (FREE - 1500 req/day)")
 
     def call_llm(self, prompt: str) -> str:
@@ -82,10 +80,20 @@ class SmartAgent:
                 "maxOutputTokens": 1024,
             }
         }
-        response = requests.post(self.api_url, json=payload, timeout=60)
-        response.raise_for_status()
-        data = response.json()
-        return data["candidates"][0]["content"]["parts"][0]["text"].strip()
+        # Built-in retry mechanism for Rate Limiting
+        for attempt in range(3):
+            try:
+                response = requests.post(self.api_url, json=payload, timeout=60)
+                response.raise_for_status()
+                data = response.json()
+                return data["candidates"][0]["content"]["parts"][0]["text"].strip()
+            except requests.exceptions.HTTPError as e:
+                if response.status_code == 429:
+                    print(f"Rate limited (429)! Waiting 10 seconds... (Attempt {attempt+1}/3)")
+                    time.sleep(10)
+                else:
+                    raise e
+        raise Exception("Failed to call LLM after 3 attempts due to rate limiting.")
 
     def __call__(self, question: str) -> str:
         print(f"\nQuestion: {question[:100]}...")
@@ -98,9 +106,10 @@ WIKIPEDIA: <query>
 PYTHON: <code>
 
 After gathering enough info, give your final answer as:
-ANSWER: <your answer>
+ANSWER: <your exact short answer>
 
 Rules for the answer:
+- DO NOT wrap the answer in "FINAL ANSWER: " or any other text. Output strictly "ANSWER: " followed by the exact answer string.
 - Numbers only (no units unless asked, no commas in numbers)
 - Short phrases (no articles like a/the, no abbreviations for proper nouns)
 - Comma-separated list if multiple items needed
@@ -109,6 +118,9 @@ Rules for the answer:
         conversation = f"{system}\n\nQuestion: {question}"
 
         for iteration in range(6):
+            # Slowing down the ReAct loop to respect Gemini's 15 RPM Free Tier limit
+            time.sleep(4.5)
+
             response = self.call_llm(conversation)
             print(f"  LLM [{iteration}]: {response[:200]}")
 
@@ -116,27 +128,27 @@ Rules for the answer:
             answer_match = re.search(r'ANSWER:\s*(.+?)(?:\n|$)', response, re.IGNORECASE)
             if answer_match:
                 answer = answer_match.group(1).strip()
-                print(f"  Final answer: {answer}")
+                print(f"   {answer}")
                 return answer
 
             tool_result = None
 
             search_match = re.search(r'SEARCH:\s*(.+?)(?:\n|$)', response)
+            wiki_match = re.search(r'WIKIPEDIA:\s*(.+?)(?:\n|$)', response)
+            python_match = re.search(r'PYTHON:\s*```(?:python)?\n?(.*?)```', response, re.DOTALL)
+            if not python_match:
+                python_match = re.search(r'PYTHON:\s*(.+?)(?:\nSEARCH|\nWIKIPEDIA|\nANSWER|$)', response, re.DOTALL)
+
+            # Execution logic changed to 'elif' to avoid overlapping tool executions
             if search_match:
                 query = search_match.group(1).strip()
                 print(f"  Tool: web_search({query})")
                 tool_result = f"Search results for '{query}':\n{web_search(query)}"
-
-            wiki_match = re.search(r'WIKIPEDIA:\s*(.+?)(?:\n|$)', response)
-            if wiki_match:
+            elif wiki_match:
                 query = wiki_match.group(1).strip()
                 print(f"  Tool: wikipedia({query})")
                 tool_result = f"Wikipedia results for '{query}':\n{wikipedia_search(query)}"
-
-            python_match = re.search(r'PYTHON:\s*```(?:python)?\n?(.*?)```', response, re.DOTALL)
-            if not python_match:
-                python_match = re.search(r'PYTHON:\s*(.+?)(?:\nSEARCH|\nWIKIPEDIA|\nANSWER|$)', response, re.DOTALL)
-            if python_match:
+            elif python_match:
                 code = python_match.group(1).strip()
                 print(f"  Tool: python({code[:50]})")
                 tool_result = f"Python output:\n{run_python(code)}"
@@ -146,7 +158,7 @@ Rules for the answer:
             else:
                 conversation += f"\n\nAssistant: {response}\n\nProvide your final answer as: ANSWER: <answer>"
 
-        # Final attempt
+        # Final attempt fallback
         conversation += "\n\nGive only the final answer as: ANSWER: <answer>"
         last = self.call_llm(conversation)
         answer_match = re.search(r'ANSWER:\s*(.+?)(?:\n|$)', last, re.IGNORECASE)
@@ -199,7 +211,9 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
         except Exception as e:
             print(f"Error on task {task_id}: {e}")
             results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"ERROR: {e}"})
-        import time; time.sleep(7)  # avoid rate limiting (free tier: 15 req/min)
+        
+        # Pausing between questions to ensure global RPM limit safety
+        time.sleep(5)
 
     if not answers_payload:
         return "Agent did not produce any answers.", pd.DataFrame(results_log)