Update app.py

app.py

@@ -232,6 +232,20 @@
 #
 # =================================================================================================
 
+# 
+
+######################
+# =================================================================================================
+#  ✅ --- ✅  FINAL ASSESSMENT AGENT - V5 (GPT-4o & PDF Support) ✅ --- ✅
+# =================================================================================================
+#
+#  Instructions:
+#  1. Add OPENAI_API_KEY, TAVILY_API_KEY, and GROQ_API_KEY to your HF Space secrets.
+#  2. Update your requirements.txt to include `langchain-openai` and `pypdf`.
+#  3. This version uses the GPT-4o model for superior reasoning and can read PDFs.
+#
+# =================================================================================================
+
 import os
 import io
 import requests
@@ -244,10 +258,11 @@ import operator
 # --- LangChain & LangGraph Imports ---
 from langchain_core.messages import BaseMessage, HumanMessage, ToolMessage, AIMessage, SystemMessage
 from langchain_core.tools import tool
-from langchain_groq import ChatGroq
+from langchain_openai import ChatOpenAI # <-- Import OpenAI
 from langgraph.graph import StateGraph, END
 from langgraph.prebuilt import ToolNode
-from tavily import TavilyClient # <-- Import Tavily
+from tavily import TavilyClient
+import pypdf # <-- Import PDF library
 
 # (Keep Constants as is)
 # --- Constants ---
@@ -255,7 +270,7 @@ DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 FILES_DIR = "./files"
 os.makedirs(FILES_DIR, exist_ok=True)
 
-# --- The new, stricter System Prompt ---
+# --- System Prompt (Unchanged, it's already strong) ---
 AGENT_SYSTEM_PROMPT = """You are a world-class AI agent, specialized in solving complex problems from the GAIA benchmark.
 
 Your task is to analyze the user's question, think step-by-step, and use the provided tools to find the correct answer.
@@ -278,24 +293,19 @@ Think, use your tools, and then provide ONLY the final, precise answer.
 
 #
 # ================================================================================================
-#  ✅ 1. DEFINE THE AGENT'S TOOLS (NOW WITH TAVILY)
+#  ✅ 1. DEFINE THE AGENT'S UPGRADED TOOLS
 # ================================================================================================
 #
-# Initialize the Tavily client. It will automatically use the TAVILY_API_KEY from secrets.
 tavily = TavilyClient(api_key=os.getenv("TAVILY_API_KEY"))
 
 @tool
 def tavily_search(query: str) -> str:
     """
     Uses the Tavily Search API to find information on the web.
-    Tavily is optimized for AI agents and provides clean, summarized results.
-    Use this for any questions that require current, factual, or web-based information.
     """
     print(f"--- Calling Tavily Search Tool with query: {query} ---")
     try:
-        # Calling the search method with the query
         result = tavily.search(query=query, search_depth="advanced")
-        # Returning the content of the search results
         return f"Search results for '{query}':\n" + "\n".join([f"- {r['content']}" for r in result['results']])
     except Exception as e:
         return f"Error during Tavily search: {e}"
@@ -303,7 +313,8 @@ def tavily_search(query: str) -> str:
 @tool
 def read_file(url: str) -> str:
     """
-    Downloads a file from a given URL and returns its content.
+    Downloads a file from a given URL, saves it locally, and returns its content.
+    It can handle both plain text files and PDF files.
     Use this tool when a question provides a URL to a file that needs to be read.
     """
     print(f"--- Calling Read File Tool with URL: {url} ---")
@@ -311,14 +322,31 @@ def read_file(url: str) -> str:
         filename = os.path.join(FILES_DIR, os.path.basename(url))
         response = requests.get(url)
         response.raise_for_status()
+        
         with open(filename, 'wb') as f:
             f.write(response.content)
-        try:
-            with open(filename, 'r', encoding='utf-8') as f:
-                content = f.read()
-            return f"Successfully read file '{filename}'. Content:\n\n{content}"
-        except UnicodeDecodeError:
-            return f"Successfully downloaded binary file '{filename}'. Cannot display content."
+            
+        # Check if the file is a PDF
+        if url.lower().endswith('.pdf'):
+            print(f"--- File identified as PDF. Reading with pypdf. ---")
+            try:
+                pdf_reader = pypdf.PdfReader(filename)
+                content = ""
+                for page in pdf_reader.pages:
+                    content += page.extract_text()
+                return f"Successfully read PDF file '{filename}'. Content:\n\n{content}"
+            except Exception as e:
+                return f"Error reading PDF file: {e}"
+        else:
+            # Assume it's a text file
+            print(f"--- File identified as text. Reading normally. ---")
+            try:
+                with open(filename, 'r', encoding='utf-8') as f:
+                    content = f.read()
+                return f"Successfully read text file '{filename}'. Content:\n\n{content}"
+            except UnicodeDecodeError:
+                return f"Successfully downloaded binary file '{filename}'. Cannot display content as text."
+
     except requests.exceptions.RequestException as e:
         return f"Error downloading or reading file: {e}"
 
@@ -326,8 +354,6 @@ def read_file(url: str) -> str:
 def python_interpreter(code: str) -> str:
     """
     Executes a given string of Python code and returns the output from stdout.
-    Use this for complex calculations, data manipulation, or any task that can be solved with code.
-    Make sure to use a print() statement to capture the output.
     """
     print(f"--- Calling Python Interpreter Tool with code:\n{code} ---")
     output_buffer = io.StringIO()
@@ -340,88 +366,70 @@ def python_interpreter(code: str) -> str:
 
 #
 # ================================================================================================
-#  ✅ 2. CONFIGURE AND BUILD THE AGENT GRAPH
+#  ✅ 2. CONFIGURE AND BUILD THE AGENT GRAPH (NOW WITH GPT-4o)
 # ================================================================================================
 #
-# This section is now self-contained to be called for each new agent instance.
-#
-
 class AgentState(TypedDict):
     messages: Annotated[List[BaseMessage], operator.add]
 
 def build_agent_graph():
     """Builds the LangGraph agent."""
     tools = [tavily_search, read_file, python_interpreter]
-    llm = ChatGroq(model="llama3-70b-8192", temperature=0)
+    
+    # --->>> THE BRAIN UPGRADE: Using GPT-4o <<<---
+    # It will use the OPENAI_API_KEY from your secrets.
+    llm = ChatOpenAI(model="gpt-4o", temperature=0)
+    
     llm_with_tools = llm.bind_tools(tools)
 
     def call_model(state: AgentState) -> dict:
-        print("--- Calling LLM ---")
         messages = state['messages']
         response = llm_with_tools.invoke(messages)
         return {"messages": [response]}
 
     def should_continue(state: AgentState) -> str:
-        last_message = state['messages'][-1]
-        if last_message.tool_calls:
-            return "action"
-        else:
-            return "end"
+        return "action" if state['messages'][-1].tool_calls else "end"
 
     tool_node = ToolNode(tools)
     workflow = StateGraph(AgentState)
     workflow.add_node("agent", call_model)
     workflow.add_node("action", tool_node)
     workflow.set_entry_point("agent")
-    workflow.add_conditional_edges(
-        "agent",
-        should_continue,
-        {"action": "action", "end": END}
-    )
+    workflow.add_conditional_edges("agent", should_continue, {"action": "action", "end": END})
     workflow.add_edge('action', 'agent')
     return workflow.compile()
 
 #
 # ================================================================================================
-#  ✅ 3. CREATE THE AGENT CLASS THAT THE TEMPLATE USES
+#  ✅ 3. AGENT CLASS AND EVALUATION LOGIC (Unchanged)
 # ================================================================================================
 #
 class GaiaAgent:
     def __init__(self):
-        print("GaiaAgent initialized. Building fresh graph...")
+        print("GaiaAgent initialized. Building fresh GPT-4o graph...")
         self.agent_app = build_agent_graph()
 
     def __call__(self, question: str) -> str:
         print(f"\n{'='*60}\nAgent received question: {question[:100]}...\n{'='*60}")
-        
         initial_input = {
             "messages": [
                 SystemMessage(content=AGENT_SYSTEM_PROMPT),
                 HumanMessage(content=question)
             ]
         }
-        
         final_state = None
         for i, step in enumerate(self.agent_app.stream(initial_input, {"recursion_limit": 15})):
-            if i == 0:
-                print("--- Starting Agentic Loop ---")
+            if i == 0: print("--- Starting Agentic Loop ---")
             final_state = step
 
         final_answer_message = final_state['agent']['messages'][-1]
         final_answer = str(final_answer_message.content).strip()
-        
         print(f"\n--- Agent finished. Final Answer: {final_answer} ---\n")
         return final_answer
 
-#
-# ================================================================================================
-#  -- EVALUATION LOGIC - CRITICAL FIX APPLIED --
-# ================================================================================================
-
 def run_and_submit_all( profile: gr.OAuthProfile | None):
     space_id = os.getenv("SPACE_ID")
-    if not profile:
-        return "Please Login to Hugging Face with the button.", None
+    if not profile: return "Please Login to Hugging Face with the button.", None
     username = f"{profile.username}"
     print(f"User logged in: {username}")
 
@@ -443,16 +451,11 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
     answers_payload = []
     print(f"Running agent on {len(questions_data)} questions...")
 
-    #
-    # --->>> CRITICAL FIX: Instantiate a NEW agent for EACH question <<<---
-    #
     for item in questions_data:
         task_id = item.get("task_id")
         question_text = item.get("question")
-        if not task_id or question_text is None:
-            continue
+        if not task_id or question_text is None: continue
         try:
-            # A new, clean agent is created here to prevent state leakage.
             agent = GaiaAgent()
             submitted_answer = agent(question_text)
             answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
@@ -467,7 +470,7 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
     submission_data = {"username": username.strip(), "agent_code": agent_code, "answers": answers_payload}
     print(f"Submitting {len(answers_payload)} answers to: {submit_url}")
     try:
-        response = requests.post(submit_url, json=submission_data, timeout=60)
+        response = requests.post(submit_url, json=submission_data, timeout=90) # Increased timeout for OpenAI
         response.raise_for_status()
         result_data = response.json()
         final_status = (
@@ -477,35 +480,29 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
             f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')} correct)\n"
             f"Message: {result_data.get('message', 'No message received.')}"
         )
-        print("Submission successful.")
         results_df = pd.DataFrame(results_log)
         return final_status, results_df
     except Exception as e:
         status_message = f"An unexpected error occurred during submission: {e}"
-        print(status_message)
         results_df = pd.DataFrame(results_log)
         return status_message, results_df
 
-
-# --- Gradio Interface (No Changes Needed) ---
+# --- Gradio Interface ---
 with gr.Blocks() as demo:
-    gr.Markdown("# GAIA Agent Final Assessment (V4 - State Fixed)")
+    gr.Markdown("# GAIA Agent Final Assessment (V5 - GPT-4o & PDF)")
     gr.Markdown(
         """
-        **Instructor's Note:** This version fixes the critical state-leakage bug and uses the Tavily Search API for better results.
-        1.  Ensure `GROQ_API_KEY` and `TAVILY_API_KEY` are set in secrets.
-        2.  Ensure `requirements.txt` includes `tavily-python`.
-        3.  Log in and run the evaluation. Let's see that score jump!
+        **Instructor's Note:** This is the final version. It uses GPT-4o for SOTA reasoning and can now read PDFs.
+        1.  Ensure `OPENAI_API_KEY` and `TAVILY_API_KEY` are set.
+        2.  Ensure `requirements.txt` is updated.
+        3.  Good luck! Let's get that certificate.
         """
     )
     gr.LoginButton()
     run_button = gr.Button("Run Evaluation & Submit All Answers")
     status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
     results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True)
-    run_button.click(
-        fn=run_and_submit_all,
-        outputs=[status_output, results_table]
-    )
+    run_button.click(fn=run_and_submit_all, outputs=[status_output, results_table])
 
 if __name__ == "__main__":
     print("\n" + "-"*30 + " App Starting " + "-"*30)