在 CrewAI 和 LangChain 中处理用户输入和脚本功能问题时遇到问题

Question

我正在使用 CrewAI 和 LangChain 开发一个用于面试模拟的脚本，其中我让代理根据用户响应生成和评估问题。但是，我在正确处理用户输入时遇到问题，并且脚本无法按预期运行。

代码片段：

import os
from crewai import Agent, Task, Crew, Process
from langchain_google_genai import ChatGoogleGenerativeAI
from dotenv import load_dotenv

load_dotenv()
gemini_llm = ChatGoogleGenerativeAI(model="gemini-pro", temperature=0)

# Define agents
question_generator = Agent(
    role='Question Generator Agent',
    goal='Generate initial and follow-up questions based on user responses and behavioral analysis.',
    verbose=True,
    memory=True,
    backstory=(
        "Expert in behavioral analysis and interview techniques, always ready with probing and engaging questions."
    ),
    llm=gemini_llm,
)

evaluator = Agent(
    role='Evaluator Agent',
    goal='Evaluate the user’s responses for relevance and sufficiency.',
    verbose=True,
    memory=True,
    backstory=(
        "Experienced evaluator with a keen eye for detail, ensuring responses meet the required criteria."
    ),
    llm=gemini_llm
)

# Define tasks
initial_question_task = Task(
    description="Generate an initial question for the user based on behavioral analysis.",
    expected_output='A single question for the user to respond to.',
    agent=question_generator,
)

evaluate_response_task = Task(
    description="Evaluate the user's response for relevance and sufficiency.",
    expected_output='Feedback on whether the response is satisfactory or if a follow-up question is needed.',
    agent=evaluator,
)

follow_up_question_task = Task(
    description="Generate a follow-up question based on the user's previous response and the evaluator's feedback.",
    expected_output='A follow-up question that probes deeper into the user\'s previous answer.',
    agent=question_generator,
)

# Define the crew and process
crew = Crew(
    agents=[question_generator, evaluator],
    tasks=[initial_question_task, evaluate_response_task, follow_up_question_task],
    process=Process.sequential
)

# Function to run the interview process
def run_interview():
    for question_number in range(10):
        print(f"\nQuestion {question_number + 1}:")
        
        # Generate the initial question
        crew.process = Process.sequential
        result = crew.kickoff(inputs={'task_name': 'initial_question_task'})
        initial_question = result['output']
        print(initial_question)
        
        user_response = input("Your response: ")
        
        # Evaluate the response
        crew.process = Process.sequential
        result = crew.kickoff(inputs={'task_name': 'evaluate_response_task', 'response': user_response})
        evaluator_feedback = result['output']
        
        if "satisfactory" in evaluator_feedback:
            print("Response is satisfactory.")
            continue
        
        # Handle follow-up questions
        for follow_up_number in range(2):
            crew.process = Process.sequential
            result = crew.kickoff(inputs={'task_name': 'follow_up_question_task', 'previous_response': user_response})
            follow_up_question = result['output']
            
            print(f"Follow-up question {follow_up_number + 1}: {follow_up_question}")
            user_response = input("Your response: ")

            crew.process = Process.sequential
            result = crew.kickoff(inputs={'task_name': 'evaluate_response_task', 'response': user_response})
            evaluator_feedback = result['output']
            
            if "satisfactory" in evaluator_feedback:
                print("Response is satisfactory.")
                break
            else:
                print("Response is not satisfactory. Generating another follow-up question...")

# Start the interview process
run_interview()

遇到的问题：

在面试模拟过程中，脚本无法正确处理用户输入。具体来说，在生成初始问题并收到用户响应后，它无法正确评估响应或根据评估者的反馈生成后续问题。相反，它会抛出错误或提供不正确的输出。

预期结果：

我希望脚本能够顺利地生成问题、评估回答并根据评估者的反馈按顺序处理后续问题。

我尝试过的：

1. 确保正确设置 CrewAI 代理和任务。
2. 验证每个任务的输入参数和预期输出。
3. 调试处理用户输入和任务执行流程的潜在问题。

Answer 1

您是如何解决这个问题的？