Upload 9 files

.gitattributes

@@ -1,35 +1,35 @@
-*.7z filter=lfs diff=lfs merge=lfs -text
-*.arrow filter=lfs diff=lfs merge=lfs -text
-*.bin filter=lfs diff=lfs merge=lfs -text
-*.bz2 filter=lfs diff=lfs merge=lfs -text
-*.ckpt filter=lfs diff=lfs merge=lfs -text
-*.ftz filter=lfs diff=lfs merge=lfs -text
-*.gz filter=lfs diff=lfs merge=lfs -text
-*.h5 filter=lfs diff=lfs merge=lfs -text
-*.joblib filter=lfs diff=lfs merge=lfs -text
-*.lfs.* filter=lfs diff=lfs merge=lfs -text
-*.mlmodel filter=lfs diff=lfs merge=lfs -text
-*.model filter=lfs diff=lfs merge=lfs -text
-*.msgpack filter=lfs diff=lfs merge=lfs -text
-*.npy filter=lfs diff=lfs merge=lfs -text
-*.npz filter=lfs diff=lfs merge=lfs -text
-*.onnx filter=lfs diff=lfs merge=lfs -text
-*.ot filter=lfs diff=lfs merge=lfs -text
-*.parquet filter=lfs diff=lfs merge=lfs -text
-*.pb filter=lfs diff=lfs merge=lfs -text
-*.pickle filter=lfs diff=lfs merge=lfs -text
-*.pkl filter=lfs diff=lfs merge=lfs -text
-*.pt filter=lfs diff=lfs merge=lfs -text
-*.pth filter=lfs diff=lfs merge=lfs -text
-*.rar filter=lfs diff=lfs merge=lfs -text
-*.safetensors filter=lfs diff=lfs merge=lfs -text
-saved_model/**/* filter=lfs diff=lfs merge=lfs -text
-*.tar.* filter=lfs diff=lfs merge=lfs -text
-*.tar filter=lfs diff=lfs merge=lfs -text
-*.tflite filter=lfs diff=lfs merge=lfs -text
-*.tgz filter=lfs diff=lfs merge=lfs -text
-*.wasm filter=lfs diff=lfs merge=lfs -text
-*.xz filter=lfs diff=lfs merge=lfs -text
-*.zip filter=lfs diff=lfs merge=lfs -text
-*.zst filter=lfs diff=lfs merge=lfs -text
-*tfevents* filter=lfs diff=lfs merge=lfs -text
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1,4 @@
+__pycache__
+.DS_Store
+tmp*
+.env

README.md

@@ -1,13 +1,15 @@
----
-title: Agent Gaia
-emoji: 🐠
-colorFrom: gray
-colorTo: red
-sdk: gradio
-sdk_version: 5.30.0
-app_file: app.py
-pinned: false
-short_description: Agents Course Final Assignment
----
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+---
+title: Agents Course Final Assignment
+emoji: 🕵🏻‍♂️
+colorFrom: indigo
+colorTo: indigo
+sdk: gradio
+sdk_version: 5.25.2
+app_file: app.py
+pinned: false
+hf_oauth: true
+# optional, default duration is 8 hours/480 minutes. Max duration is 30 days/43200 minutes.
+hf_oauth_expiration_minutes: 480
+---
+
+Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -0,0 +1,215 @@
+import os
+import gradio as gr
+import requests
+import pandas as pd
+from time import sleep
+from tqdm import tqdm
+
+# (Keep Constants as is)
+# --- Constants ---
+DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
+
+
+# --- Set framework to "langgraph" ---
+FRAMEWORK = 'langgraph'
+
+async def run_and_submit_all(profile: gr.OAuthProfile | None):
+    """
+    Fetches all questions, runs the BasicAgent on them, submits all answers,
+    and displays the results.
+    """
+    # --- Determine HF Space Runtime URL and Repo URL ---
+    space_id = os.getenv("SPACE_ID") # Get the SPACE_ID for sending link to the code
+
+    if profile:
+        username= f"{profile.username}"
+        print(f"User logged in: {username}")
+    else:
+        print("User not logged in.")
+        return "Please Login to Hugging Face with the button.", None
+
+    api_url = DEFAULT_API_URL
+    questions_url = f"{api_url}/questions"
+    submit_url = f"{api_url}/submit"
+
+    # 1. Instantiate Agent (modify this part to create your agent)
+    try:
+        if FRAMEWORK == 'langgraph':
+            from langgraph_dir.agent import LangGraphAgent
+            agent = LangGraphAgent(model_name="gpt-4.1-mini")
+        else:
+            raise AttributeError(
+                f"Set FRAMEWORK = 'langgraph', received: '{FRAMEWORK}'")
+    except Exception as e:
+        print(f"Error instantiating agent: {e}")
+        return f"Error initializing agent: {e}", None
+    # In the case of an app running as a hugging Face space, this link points toward your codebase (useful for others so please keep it public)
+    agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
+    print(agent_code)
+
+    # 2. Fetch Questions
+    print(f"Fetching questions from: {questions_url}")
+    try:
+        response = requests.get(questions_url, timeout=15)
+        response.raise_for_status()
+        questions_data = response.json()
+        if not questions_data:
+             print("Fetched questions list is empty.")
+             return "Fetched questions list is empty or invalid format.", None
+        print(f"Fetched {len(questions_data)} questions.")
+    except requests.exceptions.RequestException as e:
+        print(f"Error fetching questions: {e}")
+        return f"Error fetching questions: {e}", None
+    except requests.exceptions.JSONDecodeError as e:
+        print(f"Error decoding JSON response from questions endpoint: {e}")
+        print(f"Response text: {response.text[:500]}")
+        return f"Error decoding server response for questions: {e}", None
+    except Exception as e:
+        print(f"An unexpected error occurred fetching questions: {e}")
+        return f"An unexpected error occurred fetching questions: {e}", None
+
+    # 3. Run your Agent
+    results_log = []
+    answers_payload = []
+    print(f"Running agent on {len(questions_data)} questions...")
+    for item in questions_data:
+        task_id = item.get("task_id")
+        question_text = item.get("question")
+        file_name = item.get("file_name")
+        if not task_id or question_text is None:
+            print(f"Skipping item with missing task_id or question: {item}")
+            continue
+        try:
+            if file_name:
+                # add the URL of the data source to the question (so that the agent can deal with it)
+                file_url = f"{DEFAULT_API_URL}/files/{task_id}"
+                question_text += f'\nFile URL: "{file_url}"'
+                # get the extension of the file to help the agent
+                try:
+                    ext = file_name.split('.')[-1]
+                    question_text += f" (.{ext} file)"
+                except:
+                    pass
+
+            # call the agent
+            submitted_answer = agent(question_text)
+                
+            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})
+
+            # wait 1 minute before next call to avoid reaching limit of token per minute (TPM)
+            print('\n\n-> Sleeping for 1 minute to avoid reaching limit of token per minute (TPM)')
+            for _ in tqdm(range(60)): # tqdm to see time we have to wait
+                sleep(1)
+        except Exception as e:
+             print(f"Error running agent on task {task_id}: {e}")
+             results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"AGENT ERROR: {e}"})
+
+    if not answers_payload:
+        print("Agent did not produce any answers to submit.")
+        return "Agent did not produce any answers to submit.", pd.DataFrame(results_log)
+
+    # 4. Prepare Submission
+    submission_data = {"username": username.strip(), "agent_code": agent_code, "answers": answers_payload}
+    status_update = f"Agent finished. Submitting {len(answers_payload)} answers for user '{username}'..."
+    print(status_update)
+
+    # 5. Submit
+    print(f"Submitting {len(answers_payload)} answers to: {submit_url}")
+    try:
+        response = requests.post(submit_url, json=submission_data, timeout=60)
+        response.raise_for_status()
+        result_data = response.json()
+        final_status = (
+            f"Submission Successful!\n"
+            f"User: {result_data.get('username')}\n"
+            f"Overall Score: {result_data.get('score', 'N/A')}% "
+            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 requests.exceptions.HTTPError as e:
+        error_detail = f"Server responded with status {e.response.status_code}."
+        try:
+            error_json = e.response.json()
+            error_detail += f" Detail: {error_json.get('detail', e.response.text)}"
+        except requests.exceptions.JSONDecodeError:
+            error_detail += f" Response: {e.response.text[:500]}"
+        status_message = f"Submission Failed: {error_detail}"
+        print(status_message)
+        results_df = pd.DataFrame(results_log)
+        return status_message, results_df
+    except requests.exceptions.Timeout:
+        status_message = "Submission Failed: The request timed out."
+        print(status_message)
+        results_df = pd.DataFrame(results_log)
+        return status_message, results_df
+    except requests.exceptions.RequestException as e:
+        status_message = f"Submission Failed: Network error - {e}"
+        print(status_message)
+        results_df = pd.DataFrame(results_log)
+        return status_message, 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
+
+
+# --- Build Gradio Interface using Blocks ---
+with gr.Blocks() as demo:
+    gr.Markdown("# Basic Agent Evaluation Runner")
+    gr.Markdown(
+        """
+        **Instructions:**
+
+        1.  Please clone this space, then modify the code to define your agent's logic, the tools, the necessary packages, etc ...
+        2.  Log in to your Hugging Face account using the button below. This uses your HF username for submission.
+        3.  Click 'Run Evaluation & Submit All Answers' to fetch questions, run your agent, submit answers, and see the score.
+
+        ---
+        **Disclaimers:**
+        Once clicking on the "submit button, it can take quite some time ( this is the time for the agent to go through all the questions).
+        This space provides a basic setup and is intentionally sub-optimal to encourage you to develop your own, more robust solution. For instance for the delay process of the submit button, a solution could be to cache the answers and submit in a seperate action or even to answer the questions in async.
+        """
+    )
+
+    gr.LoginButton()
+
+    run_button = gr.Button("Run Evaluation & Submit All Answers")
+
+    status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
+    # Removed max_rows=10 from DataFrame constructor
+    results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True)
+
+    run_button.click(
+        fn=run_and_submit_all,
+        outputs=[status_output, results_table]
+    )
+
+if __name__ == "__main__":
+    print("\n" + "-"*30 + " App Starting " + "-"*30)
+
+    # Check for SPACE_HOST and SPACE_ID at startup for information
+    space_host_startup = os.getenv("SPACE_HOST")
+    space_id_startup = os.getenv("SPACE_ID") # Get SPACE_ID at startup
+
+    if space_host_startup:
+        print(f"✅ SPACE_HOST found: {space_host_startup}")
+        print(f"   Runtime URL should be: https://{space_host_startup}.hf.space")
+    else:
+        print("ℹ️  SPACE_HOST environment variable not found (running locally?).")
+
+    if space_id_startup: # Print repo URLs if SPACE_ID is found
+        print(f"✅ SPACE_ID found: {space_id_startup}")
+        print(f"   Repo URL: https://huggingface.co/spaces/{space_id_startup}")
+        print(f"   Repo Tree URL: https://huggingface.co/spaces/{space_id_startup}/tree/main")
+    else:
+        print("ℹ️  SPACE_ID environment variable not found (running locally?). Repo URL cannot be determined.")
+
+    print("-"*(60 + len(" App Starting ")) + "\n")
+
+    print("Launching Gradio Interface for Basic Agent Evaluation...")
+    demo.launch(debug=True, share=False)

get_real_rank.py

@@ -0,0 +1,27 @@
+# code to get its "true" rank on the agent course challenge
+# - remove all the users that used the code space of someone else (won't work for the forks)
+# - set the same rank for all the people with same score
+
+import pandas as pd
+from datasets import load_dataset
+
+username = 'ascythe'
+
+dataset = load_dataset('agents-course/unit4-students-scores')
+df = pd.DataFrame(dataset['train'])
+df.sort_values('score', ascending=False, inplace=True)
+
+# keep only users that submitted from their own code space or than ran locally (with None in code space)
+df['to_keep'] = df.apply(lambda row: row['username'] in row['code'] or 'None' in row['code'], axis=1)
+df = df[df['to_keep'] == True]
+
+# compute rank (all users with same score have the same rank)
+ranks_to_add = 0
+for i, score in enumerate(df['score'].unique()):
+    df.loc[df['score'] == score, 'rank'] = ranks_to_add + i + 1
+    ranks_to_add += len(df[df['score'] == score]) - 1
+
+# find `username` rank
+rank = int(df[df['username'] == username]['rank'].values[0])
+total = len(df['code'].unique())
+print(f"{username} rank: {rank}/{total} (top {rank/total*100:.1f}%)")

langgraph_dir/agent.py

@@ -0,0 +1,156 @@
+# env variable needed: OPENAI_API_KEY, BRAVE_SEARCH_API_KEY
+
+import os
+import json
+from dotenv import load_dotenv
+
+from typing import Literal
+from langchain_openai import ChatOpenAI
+from langgraph.graph import MessagesState
+from langchain_core.messages import SystemMessage, HumanMessage, ToolMessage
+from langgraph.graph import StateGraph, START, END
+from langchain_community.tools import BraveSearch, WikipediaQueryRun
+from langchain_community.utilities import WikipediaAPIWrapper
+
+from .prompt import system_prompt
+from .custom_tools import (calculator_tool, web_search, query_image, python_repl,
+    get_webdoc_content, get_website_content, extract_answer_from_content, 
+    transcribe_audio, get_youtube_transcript, generate_table_from_data, check_commutative)
+
+load_dotenv()
+# get API key from openai, and then secure the OpenAI API key in env
+openai_api_key = os.environ['OPENAI_API_KEY']
+
+class LangGraphAgent:
+    def __init__(self,
+                 model_name="gpt-4.1-mini",
+                 show_tools_desc=True,
+                 show_prompt=True):
+
+        # =========== LLM definition ===========
+        llm = ChatOpenAI(model=model_name, temperature=0, openai_api_key=openai_api_key)
+        print(f"LangGraphAgent initialized with model \"{model_name}\"")
+
+        # =========== Augment the LLM with tools ===========
+        community_tools = [
+            BraveSearch.from_api_key(   # Web search (more performant than DuckDuckGo)
+                api_key=os.getenv("BRAVE_SEARCH_API_KEY"), # needs BRAVE_SEARCH_API_KEY in env
+                search_kwargs={"count": 5}),
+        ]
+
+        wikipedia_tool = WikipediaQueryRun(api_wrapper=WikipediaAPIWrapper())
+
+        custom_tools = [
+            calculator_tool, # Basic math operations
+            web_search, # Web search using Tavily
+            query_image, # Ask anything about an image using a VLM
+            python_repl, # Python code interpreter
+            get_webdoc_content, # Load a web document
+            get_website_content, # Load a web page
+            extract_answer_from_content, # Extract an answer from a given content (e.g. PDF, web page)
+            transcribe_audio, # Transcribe an audio file to text
+            get_youtube_transcript, # Get the transcript of a YouTube video
+            generate_table_from_data, # Generate a table from a given data
+            check_commutative, # Analyzes a binary operation table for commutativity
+        ]
+
+        tools = [wikipedia_tool] + community_tools + custom_tools
+        tools_by_name = {tool.name: tool for tool in tools}
+        llm_with_tools = llm.bind_tools(tools)
+
+        # =========== Agent definition ===========
+
+        # Nodes
+        def llm_call(state: MessagesState):
+            """LLM decides whether to call a tool or not"""
+
+            return {
+                "messages": [
+                    llm_with_tools.invoke(
+                        [
+                            SystemMessage(
+                                content=system_prompt
+                            )
+                        ]
+                        + state["messages"]
+                    )
+                ]
+            }
+
+        def tool_node(state: dict):
+            """Performs the tool call"""
+
+            result = []
+            for tool_call in state["messages"][-1].tool_calls:
+                tool = tools_by_name[tool_call["name"]]
+                observation = tool.invoke(tool_call["args"])
+                result.append(ToolMessage(content=observation, tool_call_id=tool_call["id"]))
+            return {"messages": result}
+
+
+        # Conditional edge function to route to the tool node or end based upon whether the LLM made a tool call
+        def should_continue(state: MessagesState) -> Literal["environment", END]:
+            """Decide if we should continue the loop or stop based upon whether the LLM made a tool call"""
+
+            messages = state["messages"]
+            last_message = messages[-1]
+            # If the LLM makes a tool call, then perform an action
+            if last_message.tool_calls:
+                return "Action"
+            # Otherwise, we stop (reply to the user)
+            return END
+
+        # Build workflow
+        agent_builder = StateGraph(MessagesState)
+
+        # Add nodes
+        agent_builder.add_node("llm_call", llm_call)
+        agent_builder.add_node("environment", tool_node)
+
+        # Add edges to connect nodes
+        agent_builder.add_edge(START, "llm_call")
+        agent_builder.add_conditional_edges(
+            "llm_call",
+            should_continue,
+            {
+                # Name returned by should_continue : Name of next node to visit
+                "Action": "environment",
+                END: END,
+            },
+        )
+        agent_builder.add_edge("environment", "llm_call")
+
+        # Compile the agent
+        self.agent = agent_builder.compile()
+
+        if show_tools_desc:
+            for i, tool in enumerate(llm_with_tools.kwargs['tools']):
+                print("\n" + "="*30 + f" Tool {i+1} " + "="*30)
+                print(json.dumps(tool[tool['type']], indent=4))
+
+        if show_prompt:
+            print("\n" + "="*30 + f" System prompt " + "="*30)
+            print(system_prompt)
+
+
+    def __call__(self, question: str) -> str:
+        print("\n\n"+"*"*20)
+        print(f"Agent received question: {question}")
+        print("*"*20)
+
+        # Invoke
+        messages = [HumanMessage(content=question)]
+        messages = self.agent.invoke({"messages": messages},
+                                     {"recursion_limit": 30}) # maximum number of steps before hitting a stop condition
+        for m in messages["messages"]:
+            m.pretty_print()
+
+        # post-process the response (keep only what's after "FINAL ANSWER:" for the exact match)
+        response = str(messages["messages"][-1].content)
+        try:
+            response = response.split("FINAL ANSWER:")[-1].strip()
+        except:
+            print('Could not split response on "FINAL ANSWER:"')
+        print("\n\n"+"-"*50)
+        print(f"Agent returning with answer: {response}")
+        return response

langgraph_dir/custom_tools.py

@@ -0,0 +1,559 @@
+from langchain_core.tools import tool, Tool
+import math
+
+@tool
+def calculator_tool(expression: str) -> str:
+    """
+    Evaluate a mathematical expression.
+    """
+    # Define the restricted global and local namespace
+    safe_globals = {"__builtins__": {}}
+    
+    safe_locals = {
+        # Math functions
+        'sqrt': math.sqrt,
+        'sin': math.sin,
+        'cos': math.cos,
+        'tan': math.tan,
+        'log': math.log10,   # log base 10
+        'ln': math.log,      # natural log
+        'exp': math.exp,
+        'pow': pow,
+        
+        # Constants
+        'pi': math.pi,
+        'e': math.e,
+
+        # Built-in math utilities
+        'abs': abs,
+        'round': round,
+        'max': max,
+        'min': min,
+        'sum': sum,
+    }
+
+    try:
+        # Evaluate the expression in a restricted environment
+        result = eval(expression, safe_globals, safe_locals)
+
+        # Handle None explicitly
+        if result is None:
+            return "calculator tool produced no valid result"
+
+        # Optional: Round very small floats to avoid scientific notation
+        if isinstance(result, float) and abs(result) < 1e-9:
+            result = round(result, 10)
+
+        return str(result)
+
+    except SyntaxError as se:
+        return f"Syntax error in expression: {str(se)}"
+    except NameError as ne:
+        return f"Undefined variable or function used: {str(ne)}"
+    except ZeroDivisionError:
+        return "Error: Division by zero"
+    except Exception as e:
+        return f"Evaluation error: {str(e)}"
+
+from langchain_tavily import TavilySearch
+
+@tool
+def web_search(query: str) -> str:
+    """
+    Searches the web and returns a list of the most relevant URLs.
+    Use this FIRST for complex queries, metadata questions, or to find the right sources.
+    Then follow up with get_webdoc_content or get_website_content on the most promising URL.
+    """
+    try:
+        tavily_search = TavilySearch(
+            max_results=5,
+            topic="general",
+            search_depth="advanced",
+            include_raw_content=False,  # Just URLs and snippets
+        )
+
+        results = tavily_search.invoke(query)
+        # Format results to show URLs and brief descriptions
+        web_search_results = "Search Results:\n"
+        for i, result in enumerate(results["results"], 1):
+            web_search_results += f"{i}. {result['title']}: {result['url']}\n   {result['content'][:150]}...\n\n"
+
+        return web_search_results
+    except Exception as e:
+        return f"web_search tool error: {str(e)}"
+
+import os
+import tempfile
+import requests
+import easyocr
+from io import BytesIO
+from PIL import Image
+from openai import OpenAI
+        
+@tool
+def query_image(query: str, source: str, need_ocr: bool = True, need_reasoning: bool = False) -> str:
+    """Use ONLY to answer question about an image using a Vision Language Model.
+       NOT used to perform image processing or other tasks EXCEPT asking question about an image.
+    Args:
+        query (str): The question about the image, e.g. how many persons are on the image?
+        source (str): URL to the image
+        need_reasoning (bool): Set to True for complex query that require a reasoning model to answer properly. Set to False otherwise.
+        need_ocr (bool): If True, also extract visible text from the image. Set to False otherwise.
+    """
+
+    try:
+        # OCR Extraction (optional)
+        ocr_text = ""
+        if need_ocr:
+            try:
+                # Download image from URL
+                response = requests.get(source, stream=True, timeout=10)
+                response.raise_for_status()
+
+                # Load image into PIL
+                image = Image.open(BytesIO(response.content))
+
+                # Save to temporary file
+                with tempfile.NamedTemporaryFile(delete=False, suffix='.jpg') as tmpfile:
+                    image.save(tmpfile, format=image.format)
+                    file_to_use = tmpfile.name
+
+                # Perform OCR
+                reader = easyocr.Reader(['en'])
+                results = reader.readtext(file_to_use)
+                ocr_text = "\n".join([res[1] for res in results])
+                ocr_text = f"\n\n[OCR Extracted Text]:\n{ocr_text}"
+
+            except Exception as ocr_error:
+                ocr_text = f"\n\n[OCR Error]: {str(ocr_error)}"
+            finally:
+                # Clean up temporary file
+                if file_to_use and os.path.exists(file_to_use):
+                    os.unlink(file_to_use)
+
+        # Query Vision Language Model       
+        client = OpenAI()
+        if need_reasoning:
+            model_name = "o4-mini"
+        else:
+            model_name = "gpt-4o-mini"
+        response = client.chat.completions.create(
+            model=model_name,
+            messages=[
+                {
+                    "role": "user",
+                    "content": [
+                        {"type": "text", "text": query},
+                        {"type": "image_url", "image_url": {"url": source}},
+                    ],
+                }
+            ],
+            max_tokens=512,
+        )
+        content = response.choices[0].message.content
+
+        # Combine OCR and VLM output
+        final_response = content
+        if need_ocr and ocr_text:
+            final_response += ocr_text
+
+        return final_response
+
+    except Exception as e:
+        return f"Image query failed: {str(e)}"
+
+from pydantic import BaseModel, Field
+from e2b import Sandbox
+import re
+import os
+
+class PythonCodeInput(BaseModel):
+    code: str = Field(description="The Python code string to execute.")
+
+@tool(args_schema=PythonCodeInput)
+def python_repl(code: str) -> str:
+    """    
+    Use this to execute single or multi-line Python commands to perform tasks like:
+    sort a list in ascending or descending order, reverse input string, draw a table, photo processing, etc.
+    
+    Input should be syntactically valid Python code.
+    Make sure to include required imports in the code.
+    Always include in your code `print(...)` or `image.save(...)` to return outputs that can be seen.
+    
+    You are allowed to access internet and download files from URLs via code (e.g., using requests)
+    Avoid using any system-level commands or libraries that could harm the host system.
+    Avoid commands that require user input or block indefinitely (e.g., `input()`).
+    """
+
+    # List of forbidden patterns in code
+    FORBIDDEN_PATTERNS = [
+        r'\bimport\s+(os|sys|subprocess|shutil|socket)',
+        r'\b(eval|exec|input|open)\s*$(?=.*\w)',
+        r'\b__import__',
+        r'\bos\.',
+        r'\bsys\.',
+        r'\bsubprocess\.',
+    ]
+
+    # Step 1: Keyword-based security check
+    for pattern in FORBIDDEN_PATTERNS:
+        if re.search(pattern, code):
+            match = re.search(pattern, code).group()
+            return f"Blocked unsafe operation: {match}"
+
+    # Step 2: Create E2B sandbox
+    try:
+        with Sandbox(api_key=os.getenv("E2B_API_KEY")) as sandbox:
+            # Known mismatches: import name -> pip package name
+            import_to_pip = {
+                "PIL": "pillow",
+                "cv2": "opencv-python",
+                "yaml": "PyYAML",
+                "bs4": "beautifulsoup4",
+                "tkinter": "tk",
+            }
+
+            # Built-in modules that don't need installation
+            built_in_modules = {
+                "math", "re", "json", "csv", "os", "sys", "time", "datetime", "random", 
+                "itertools", "functools", "__future__", "collections", "pathlib", "io",
+            }
+
+            # Step 1: Extract import statements
+            import_matches = re.findall(
+                r'(?:import\s+([a-zA-Z0-9_]+)(?!\.)|\bfrom\s+([a-zA-Z0-9_]+)(?=\s+import\b))',
+                code
+            )
+            base_imports = set()
+            base_imports = set(match[0] or match[1] for match in import_matches)    # match[0] = 'import X', match[1] = 'from X import Y'
+
+            # Step 2: Determine which packages to install
+            packages_to_install = set()
+
+            for imp in base_imports:
+                # Skip known built-ins
+                if imp in built_in_modules:
+                    continue
+
+                # Use mapped name if exists, else use import name
+                package_name = import_to_pip.get(imp, imp)
+
+                # Avoid installing system-specific modules like __pycache__
+                if imp.startswith("__"):
+                    continue
+
+                packages_to_install.add(package_name)
+
+            # Step 3: Install necessary packages
+            if packages_to_install:
+                install_cmd = f"pip install {' '.join(packages_to_install)}"
+                result = sandbox.commands.run(install_cmd)
+
+                if result.stderr:
+                    return f"Failed to install packages:\n{result.stderr}"
+
+            # Step 4: Write and run the user code
+            CODE_FILE_PATH = "/tmp/code.py"
+            sandbox.files.write(CODE_FILE_PATH, code)
+
+            # Step 5: Execute the code using the new API
+            result = sandbox.commands.run(f"python {CODE_FILE_PATH}")
+            stdout = result.stdout.strip()
+            stderr = result.stderr.strip()
+
+            # Step 6: Return output
+            if stderr:
+                return f"Execution error:\n{stderr}"
+
+            return stdout or "No output"
+                        
+    except Exception as e:
+        return f"Sandbox error: {str(e)}"
+
+import requests
+from bs4 import BeautifulSoup
+from PyPDF2 import PdfReader
+from io import BytesIO
+from markdownify import markdownify
+
+@tool
+def get_webdoc_content(url: str) -> str:
+    """
+    Extracts content from PDFs or document-like URLs (academic papers, reports)
+    Can be used after web_search to get detailed information.
+    Args:
+        url (str): the URL of web page to extract the content from
+    """
+    try:
+        headers = {
+            "User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36"
+        }
+
+        response = requests.get(url, headers=headers, timeout=10)
+        response.raise_for_status()
+
+        content_type = response.headers.get('Content-Type', '')
+        
+        # PDF Handling
+        if 'application/pdf' in content_type:
+            pdf_file = BytesIO(response.content)
+            reader = PdfReader(pdf_file)
+            text = "\n".join(page.extract_text() for page in reader.pages)
+            # return f"## PDF Content from {page_url}\n\n{text[:15000]}"
+            return f"## PDF Content from {url}\n\n```\n{text[:15000]}\n```"
+
+        # HTML Document Handling
+        elif 'text/html' in content_type:
+            soup = BeautifulSoup(response.text, 'html.parser')
+            cleaned_html = soup.body or soup  # Fallback to full document
+            return markdownify(str(cleaned_html), strip=['a'])
+
+        # Fallback: Raw text extraction
+        else:
+            return f"## Raw Content from {url}\n\n{response.text[:15000]}"
+
+    except requests.exceptions.RequestException as e:
+        return f"HTTP error in get_webpage_content: {str(e)}"
+    except Exception as e:
+        return f"Unexpected error in get_webpage_content: {str(e)}"
+
+import requests
+from bs4 import BeautifulSoup
+from markdownify import markdownify
+
+@tool
+def get_website_content(url: str) -> str:
+    """
+    Extracts contents from HTML-based URLs.
+    Specializes in Wikipedia, technical documentation, and discussion pages.
+    NOT used for document-based URLs (academic papers, reports).
+    Used after web_search to get detailed information.
+    Args:
+        url (str): The URL of the web page to extract content from
+    """
+    try:
+        headers = {
+            "User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36"
+        }
+
+        response = requests.get(url, headers=headers, timeout=10)
+        response.raise_for_status()
+
+        soup = BeautifulSoup(response.text, 'html.parser')
+        # Remove non-content elements
+        for element in soup.select('script, style, footer, nav, header, aside'):
+            element.decompose()  
+
+        # Convert cleaned HTML to markdown
+        cleaned_html = str(soup.body) if soup.body else str(soup)
+        markdown_content = markdownify(cleaned_html, strip=['a'])  # Optional: strip links
+
+        return f"## Extracted Content from {url}\n\n{markdown_content[:15000]}"  # Limit length
+
+    except requests.exceptions.RequestException as e:
+        return f"HTTP error in web_content_extract: {str(e)}"
+    except Exception as e:
+        return f"Unexpected error in web_content_extract: {str(e)}"
+
+import os
+from langchain_openai import OpenAIEmbeddings
+from langchain_community.vectorstores import FAISS
+from langchain_text_splitters import RecursiveCharacterTextSplitter
+
+@tool
+def extract_answer_from_content(content: str | dict, query: str) -> str:
+    """
+    Extract relevant information from content based on user query.
+    
+    Args:
+        content (str/dict): Raw text or transcribed test from audio or structured content from any source
+        query (str): Natural language question to answer
+        
+    Returns:
+        str: Concise answer extracted from content
+    """
+    try:
+        # Normalize content format
+        if isinstance(content, dict):
+            text_content = ""
+            if "summary" in content:
+                text_content += f"SUMMARY: {content['summary']}\n\n"
+            if "infobox" in content:
+                text_content += "INFOBOX:\n"
+                for k, v in content["infobox"].items():
+                    text_content += f"{k}: {v}\n"
+                text_content += "\n"
+            if "sections" in content:
+                for section, text in content["sections"].items():
+                    text_content += f"{section}:\n{text}\n\n"
+        else:
+            text_content = content
+            
+        # Initialize OpenAI embeddings
+        embeddings = OpenAIEmbeddings(
+            openai_api_key=os.getenv("OPENAI_API_KEY"),
+            model="text-embedding-3-large"
+        )
+        
+        # Split content into manageable chunks
+        text_splitter = RecursiveCharacterTextSplitter(
+            chunk_size=500,
+            chunk_overlap=100
+        )
+        chunks = text_splitter.split_text(text_content)
+        
+        # Create vector store
+        vectorstore = FAISS.from_texts(chunks, embeddings)
+        retriever = vectorstore.as_retriever(search_kwargs={"k": 3})
+        
+        # Get most relevant content
+        relevant_docs = retriever.invoke(query)
+        combined_text = " ".join([doc.page_content for doc in relevant_docs])
+        
+        # Return relevant content with context
+        return f"Relevant information found:\n{combined_text[:1500]}"
+        
+    except Exception as e:
+        return f"Content extraction failed: {str(e)}"
+
+import os
+import requests
+from openai import OpenAI
+
+@tool
+def transcribe_audio(source: str, file_extension: str) -> str:
+    """
+    Transcribes an audio to text from local path or URL.
+    Args:
+        source (str): URL to an audio file.
+    
+    Returns:
+        str: The transcribed text, or error message.
+    """
+    # If file is not existing use download_file_from_url tool to download the file first.
+
+    client = OpenAI()
+
+    try:
+        # download the audio file
+        response = requests.get(source)
+        response.raise_for_status()
+        # write to disk
+        file_extension = file_extension.replace('.','')
+        with open(f'tmp.{file_extension}', 'wb') as file:
+            file.write(response.content)
+
+        audio_file = open(f'tmp.{file_extension}', "rb")
+        client = OpenAI()
+        transcription = client.audio.transcriptions.create(
+            model="whisper-1",
+            file=audio_file
+        )
+        return transcription.text
+
+    except Exception as e:
+        return f"Transcription error: {str(e)}"
+
+from youtube_transcript_api import YouTubeTranscriptApi
+from pytube import extract
+
+@tool
+def get_youtube_transcript(page_url: str) -> str:
+    """Get the transcript of audio component of YouTube video.
+    Use this for Youtube videos with available transcripts
+    Args:
+        page_url (str): YouTube URL of the video
+    """
+    try:
+        # Get video ID from URL
+        video_id = extract.video_id(page_url)
+        
+        # Get transcript using correct method
+        transcript = YouTubeTranscriptApi.get_transcript(video_id)
+        
+        # Return concatenated text
+        return '\n'.join([s['text'] for s in transcript])
+    
+    except Exception as e:
+        return f"get_youtube_transcript failed: {str(e)}"
+
+from tabulate import tabulate
+from typing import Dict, Any, List
+
+@tool
+def generate_table_from_data(data: List[Dict[str, Any]]) -> str:
+    """
+    Convert list of dictionaries to markdown table
+    
+    Args:
+        data (List[Dict]): List of objects with common keys
+    
+    Returns:
+        str: Markdown-formatted table
+    """
+    if not data:
+        return "No data available"
+        
+    headers = data[0].keys()
+    rows = [list(item.values()) for item in data]
+    
+    return tabulate(rows, headers=headers, tablefmt="pipe")
+
+from pydantic import BaseModel, Field
+from typing import List, Dict 
+
+class CommutativeCheckInput(BaseModel):
+    table_str: str = Field(..., description="Markdown-formatted string of the operation table (e.g., |*|a|b|c|...)")
+    elements: List[str] = Field(..., description="List of elements in the set S")
+
+@tool(args_schema=CommutativeCheckInput)
+def check_commutative(table_str: str, elements: List[str]) -> str:
+    """
+    Analyzes a binary operation table for commutativity.
+    
+    Args:
+        table_str (str): Markdown-formatted string of the operation table.
+        elements (List[str]): List of elements in the set S.
+        
+    Returns:
+        str: Comma-separated list of element pairs (e.g., "b,e") where x*y ≠ y*x.
+    """
+    
+    # Parse the table string into a 2D list
+    lines = [line.strip() for line in table_str.strip().split('\n') if line.strip()]
+    header = [cell.strip() for cell in lines[0].split('|') if cell.strip()][1:]  # Skip the first cell (operator)
+    rows = []
+    for line in lines[2:]:
+        cells = [cell.strip() for cell in line.split('|') if cell.strip()]  # Remove empty cells
+        if cells:
+            rows.append(cells)
+
+
+    # Validate that all rows have the correct number of cells
+    expected_length = len(header) + 1  # x + one for each header
+    for row in rows:
+        if len(row) < expected_length:
+            return f"Error: Row '{row[0]}' has {len(row)} cells, but expected {expected_length}."
+
+    # Build a dictionary for the operation: op[x][y] = result
+    operation: Dict[str, Dict[str, str]] = {}
+    for row in rows:
+        x = row[0]
+        operation[x] = {}
+        for i, y in enumerate(header):
+            operation[x][y] = row[i + 1]
+
+    # Check all pairs (x, y) for x*y == y*x
+    counterexamples = []
+    for x in elements:
+        for y in elements:
+            if x < y:  # Avoid redundant checks and self-comparison
+                try:
+                    xy = operation[x][y]
+                    yx = operation[y][x]
+                    if xy != yx:
+                        counterexamples.append(f"{x},{y}")
+                except KeyError as e:
+                    return f"Error: Missing data for pair ({x}, {y}) in table."
+
+    return "\n".join(counterexamples) if counterexamples else "The operation is commutative."

langgraph_dir/prompt.py

@@ -0,0 +1,11 @@
+# customized GAIA system prompt
+system_prompt = """\
+You are a general AI assistant with tools.
+I will ask you a question. Use your tools, and 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 your answer is an instruction, you need to perform that instruction too, until you give an answer that is not an instruction.
+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), and write the digits in plain text unless specified otherwise.
+If you are asked for a comma separated list, apply the above rules depending of whether the element to be put in the list is a number or a string.
+If you decide to write a code, use the python_repl tool to write code and run it.
+"""

requirements.txt

@@ -0,0 +1,23 @@
+gradio
+requests
+pandas
+tqdm
+datasets
+langchain
+langgraph
+langchain-openai
+langchain-community
+langchain-tavily
+langchain_experimental
+easyocr
+Pillow
+openai
+pydantic
+beautifulsoup4
+PyPDF2
+markdownify
+youtube-transcript-api
+pytube
+tabulate
+e2b
+duckduckgo-search