app.py

# pip install langgraph langchain-google-genai langchain-groq
from langgraph.graph import StateGraph, END
from typing import TypedDict
import requests
from langchain_google_genai import ChatGoogleGenerativeAI # or from langchain_groq import ChatGroQ

# 1. Define your model (Choose one)
llm = ChatGoogleGenerativeAI(model="gemini-1.5-flash", google_api_key="AIzaSyBkAb7nRruE-Sn-Ff9bMWhiVBvY751pqOk")
# llm = ChatGroQ(model="mixtral-8x7b-32768", groq_api_key="your_key") # Fast & cheap for experimentation

# 2. Define your agent's state. What information does it need to carry between steps?
class AgentState(TypedDict):
    question: str
    task_id: str | None  # For fetching files
    agent_response: str
    final_answer: str

# 3. Define the nodes (functions) of your graph

def fetch_question_node(state: AgentState):
    """Node 1: Calls the /random-question API to get a problem."""
    response = requests.get("https://agents-course-unit4-scoring.hf.space/random-question")
    response.raise_for_status()
    data = response.json()
    print("this is a data",data)
    return {
        "question": data["question"],
        "task_id": data.get("task_id")  # If the question requires a file, the API might provide an ID
    }

def download_file_node(state: AgentState):
    """Node 2: If a task_id is present, download the related file."""
    if not state['task_id']:
        return {"agent_response": "No file to download."}
    
    file_url = f"https://agents-course-unit4-scoring.hf.space/files/{state['task_id']}"
    file_response = requests.get(file_url)
    # Process the file (e.g., if it's a text file, read it; if it's an image, use a vision model)
    file_content = file_response.text
    enhanced_query = f"Question: {state['question']}\n\nUse the following file content to answer:\n{file_content}"
    return {"agent_response": enhanced_query}

def reason_and_answer_node(state: AgentState):
    """Node 3: The core reasoning node. Use the LLM to solve the problem."""
    # Construct a powerful prompt to encourage step-by-step thinking internally,
    # but demand a final, clean answer.
    system_prompt = """You are a highly intelligent AI agent solving a question.
    Think carefully about the problem. Your final output must be ONLY the final answer, with no additional text, explanations, or formatting.
    """
    
    prompt = f"{system_prompt}\n\nQuestion: {state['agent_response']}"
    response = llm.invoke(prompt)
    
    # Here you might need to parse the response to extract *only* the final answer.
    # For a well-instructed model, `response.content` should be just the answer.
    return {"final_answer": response.content.strip()}

# def submit_answer_node(state: AgentState):
#     """Node 4: Submit the final answer to the scoring API."""
#     payload = {
#         "question_id": state.get("question_id"), # You might need to get this from the initial fetch
#         "answer": state['final_answer']
#     }
#     submission_response = requests.post("https://agents-course-unit4-scoring.hf.space/submit", json=payload)
#     print(f"Submission Status: {submission_response.status_code}")
#     print(f"Submission Response: {submission_response.json()}")
#     return state




def submit_answer_node(state: AgentState):
    """Node 4: Submit the final answer to the scoring API."""
    
    payload = {
        "username": "your_hf_username",  # 👈 replace with your Hugging Face username
        "agent_code": "https://huggingface.co/spaces/your-username/your-space-name/tree/main",  # 👈 replace with your Space link
        "answers": [
            {
                "task_id": state.get("task_id"),  # use task_id from initial fetch
                "submitted_answer": state["final_answer"]
            }
        ]
    }

    submission_response = requests.post(
        "https://agents-course-unit4-scoring.hf.space/submit",
        json=payload
    )

    print(f"Submission Status: {submission_response.status_code}")
    print(f"Submission Response: {submission_response.json()}")
    
    return state


# 4. Build the Graph
workflow = StateGraph(AgentState)

# Add nodes
workflow.add_node("fetch_question", fetch_question_node)
workflow.add_node("download_file", download_file_node)
workflow.add_node("reason", reason_and_answer_node)
workflow.add_node("submit", submit_answer_node)

# Define the graph flow
workflow.set_entry_point("fetch_question")
workflow.add_edge("fetch_question", "download_file")
workflow.add_edge("download_file", "reason")
workflow.add_edge("reason", "submit")
workflow.add_edge("submit", END)

# Compile the graph
app = workflow.compile()