利用AI Agent在云端搭建智能HR员工助手（下）
使用提供的查询参数从知识库中检索相关文档。有关如何使用Amazon Bedrock创建知识库的更多信息，请参考官方文档。在本示例中，我们使用了一些示例HR政策文档，并同步存储到Amazon S3数据源中。大家还可以访问aws-samples Github代码库中的示例代码，以了解如何使用boto3与Amazon Bedrock的知识库进行交互。},：从知识库中检索到的文档中，提取上下文信息以供大模
·
在上篇教程中，我们展示了如何利用Amazon Bedrock的Converse API和函数调用能力来优化HR为员工提供支持，并提供了一个示例代码来展示如何集成HR系统，以实现更智能、更高效的HR流程管理。在下篇中我们将展示该方案的具体功能性实操代码。
代码功能介绍
-
retrieve(query, kbId, numberOfResults)：使用提供的查询参数从知识库中检索相关文档。有关如何使用Amazon Bedrock创建知识库的更多信息，请参考官方文档。在本示例中，我们使用了一些示例HR政策文档，并同步存储到Amazon S3数据源中。
大家还可以访问aws-samples Github代码库中的示例代码，以了解如何使用boto3与Amazon Bedrock的知识库进行交互。
def retrieve(query, kbId, numberOfResults=3):
bedrock_config = Config(connect_timeout=120, read_timeout=120, retries={'max_attempts': 0})
bedrock_agent_client = boto3.client("bedrock-agent-runtime",config=bedrock_config)
return bedrock_agent_client.retrieve(
retrievalQuery= {
'text': query
},
knowledgeBaseId=kbId,
retrievalConfiguration= {
'vectorSearchConfiguration': {
'numberOfResults': numberOfResults,
'overrideSearchType': "HYBRID"
}
}
)-
get_contexts(retrievalResults)：从知识库中检索到的文档中，提取上下文信息以供大模型进一步处理。此部分可以根据具体需求进行调整。
def get_contexts(retrievalResults):
contexts = []
#print(retrievalResults)
for retrievedResult in retrievalResults:
text = retrievedResult['content']['text']
# Remove the "Document 1: " prefix if it exists
if text.startswith("Document 1: "):
text = text[len("Document 1: "):]
contexts.append(text)
contexts_string = ', '.join(contexts)
return contexts_string-
get_pto_balance(employee_id)：根据员工ID获取其PTO剩余天数。
def get_pto_balance(employee_id):
"""Returns the PTO balance for the given employee ID."""
pto_balance = 0
if employee_id == "E1234":
pto_balance = 10
elif employee_id == "E5678":
pto_balance = 5
else:
raise ValueError(f"Invalid employee ID: {employee_id}")
return pto_balance
-
submit_pto_request(employee_id, start_date, end_date, reason)：提交员工的PTO申请。
def submit_pto_request(employee_id, start_date, end_date, reason):
"""Submits a PTO request for the given employee ID, start date, end date, and reason."""
pto_balance = get_pto_balance(employee_id)
start_date = datetime.strptime(start_date, "%Y-%m-%d")
end_date = datetime.strptime(end_date, "%Y-%m-%d")
days = (end_date - start_date).days + 1
if days > pto_balance:
raise InsufficientPTOBalanceError(f"Insufficient PTO balance for employee {employee_id}.")
# Simulating PTO request submission
return f"PTO request submitted for employee {employee_id} from {start_date.strftime('%Y-%m-%d')} to {end_date.strftime('%Y-%m-%d')} for {reason}."
-
InsufficientPTOBalanceError：一个自定义异常类，当员工尝试提交PTO请求但余额不足时会抛出此异常。
class InsufficientPTOBalanceError(Exception):
"""Raised when an employee doesn't have enough PTO balance."""
pass-
get_hr_policy(question)：根据员工提供的问题，从Amazon Bedrock的知识库中检索特定HR政策信息，并将结果传递给Amazon Bedrock模型，以生成响应并返回员工最终结果。
def get_hr_policy(question):
response_retrieve = retrieve(question, "<Your_KnowledgeBase_ID>")["retrievalResults"]
contexts = get_contexts(response_retrieve)
PROMPT_TEMPLATE = """DOCUMENT:
{context}
QUESTION:
{message}
INSTRUCTIONS:
Answer the user's QUESTION using only the DOCUMENT text above.
Keep your answer strictly grounded in the facts provided. Do not refer to the "DOCUMENT," "documents," "provided text," ,"based on.." or any similar phrases in your answer.
If the provided text contains the facts to answer the QUESTION, include all relevant details in your answer.
If the provided text doesn’t contain the facts to answer the QUESTION, respond only with "I don't know" and do not add any further information.
"""
prompt = PromptTemplate(template=PROMPT_TEMPLATE,
input_variables=["context","message"])
prompt_final = prompt.format(context=contexts,
message=question)
body = json.dumps({
"anthropic_version": "bedrock-2023-05-31",
"max_tokens": 1000,
"messages": [
{
"role": "user",
"content": [
{
"type": "text",
"text": prompt_final
}
]
}
]
})
model_id = 'anthropic.claude-3-sonnet-20240229-v1:0'
accept = 'application/json'
content_type = 'application/json'
response = bedrock_client.invoke_model(body=body, modelId=model_id, accept=accept, contentType=content_type)
response_body = json.loads(response.get('body').read())
response_text = response_body.get('content')[0]['text']
return response_text-
generate_text(bedrock_client, model_id, tool_config, input_text)：使用Amazon Bedrock模型生成文本，并触发第三方工具请求的调用。
def generate_text(bedrock_client, model_id, tool_config, input_text):
"""Generates text using the supplied Amazon Bedrock model. If necessary,
the function handles tool use requests and sends the result to the model."""
logger.info("Generating text with model %s", model_id)
# Create the initial message from the user input.
messages = [{
"role": "user",
"content": [{"text": input_text}]
}]
response = bedrock_client.converse(
modelId=model_id,
messages=messages,
toolConfig=tool_config
)
output_message = response['output']['message']
messages.append(output_message)
stop_reason = response['stopReason']
if stop_reason == 'tool_use':
# Tool use requested. Call the tool and send the result to the model.
tool_requests = response['output']['message']['content']
print(tool_requests)
for tool_request in tool_requests:
if 'toolUse' in tool_request:
tool = tool_request['toolUse']
logger.info("Requesting tool %s. Request: %s",
tool['name'], tool['toolUseId'])
if tool['name'] == 'pto_balance':
tool_result = {}
try:
pto_balance = get_pto_balance(tool['input']['employee_id'])
tool_result = {
"toolUseId": tool['toolUseId'],
"content": [{"json": {"pto_balance": pto_balance}}]
}
except ValueError as err:
tool_result = {
"toolUseId": tool['toolUseId'],
"content": [{"text": err.args[0]}],
"status": 'error'
}
tool_result_message = {
"role": "user",
"content": [
{
"toolResult": tool_result
}
]
}
messages.append(tool_result_message)
elif tool['name'] == 'submit_pto':
tool_result = {}
try:
pto_request_result = submit_pto_request(
tool['input']['employee_id'],
tool['input']['start_date'],
tool['input']['end_date'],
tool['input']['reason']
)
tool_result = {
"toolUseId": tool['toolUseId'],
"content": [{"text": pto_request_result}]
}
except (ValueError, InsufficientPTOBalanceError) as err:
tool_result = {
"toolUseId": tool['toolUseId'],
"content": [{"text": err.args[0]}],
"status": 'error'
}
tool_result_message = {
"role": "user",
"content": [
{
"toolResult": tool_result
}
]
}
messages.append(tool_result_message)
elif tool['name'] == 'hr_policy':
tool_result = {}
policy = get_hr_policy(input_text)
policy = {
"text": policy
}
print(json.dumps(policy, indent=4))
return
# Send the tool result to the model.
response = bedrock_client.converse(
modelId=model_id,
messages=messages,
toolConfig=tool_config
)
print(response)
output_message = response['output']['message']
# Print the final response from the model.
for content in output_message['content']:
print(json.dumps(content, indent=4))处理工具请求
generate_text函数使用converse API将用户输入发送到模型。如果模型决定调用三方工具（即stop_reason == 'tool_use'），则函数会根据工具名称处理请求：
- 对于
pto_balance，调用get_pto_balance函数，获取提供的员工ID的PTO余额，并返回余额信息。 - 对于
submit_pto，调用submit_pto_request函数，提交PTO请求，并返回执行结果。 - 对于
hr_policy，调用get_hr_policy函数，查询知识库中的HR政策，并返回检索到的政策信息。
在处理完工具调用请求后，该函数会将工具执行结果返回给模型，并通过converse API继续对话，让模型生成最终的响应。
智能助手使用与测试
更新example_questions列表，加入大家自己的HR相关问题。以下是一个示例集合：
example_questions = [
"What is the PTO balance for employee E5678?",
"Submit a PTO request for employee E1234 from 2023-07-10 to 2023-07-12 for a family vacation.",
"What is the dress code policy?",
"Can you tell me how many PTO days employee E1234 has left?",
"I want to request PTO for employee E5678 starting from 2023-08-01 to 2023-08-05 for a personal trip.",
"What are the standard working hours?",
"I need to know the remaining PTO balance for employee E9012.",
"Employee E9012 needs to take PTO from 2023-09-15 to 2023-09-20 for a medical procedure.",
"What is the overtime policy?",
"What is the policy for vacation",
"What is the policy for work from home?"
]运行脚本：
for question in example_questions:
try:
print(f"Question: {question}")
generate_text(bedrock_client, model_id, tool_config, question)
print("\n")
except ClientError as err:
message = err.response['Error']['Message']
logger.error("A client error occurred: %s", message)
print(f"A client error occurred: {message}")输出结果：大家可以看到，根据用户输入，合适的工具会被调用并执行任务。
Question: What is the PTO balance for employee E5678?
INFO:__main__:Generating text with model anthropic.claude-3-sonnet-20240229-v1:0
[{'text': 'To get the PTO balance for an employee, we can use the "pto_balance" tool:'}, {'toolUse': {'toolUseId': 'tooluse_00kL9hLpQOelFoaQjbDSlw', 'name': 'pto_balance', 'input': {'employee_id': 'E5678'}}}]
INFO:__main__:Requesting tool pto_balance. Request: tooluse_00kL9hLpQOelFoaQjbDSlw
{
"text": "The PTO balance for employee E5678 is 5 days."
}
Question: Submit a PTO request for employee E1234 from 2023-07-10 to 2023-07-12 for a family vacation.
INFO:__main__:Generating text with model anthropic.claude-3-sonnet-20240229-v1:0
[{'text': 'Here is how to submit the PTO request using the tool:'}, {'toolUse': {'toolUseId': 'tooluse_ywjLWDjfSISVP91AfBAmag', 'name': 'submit_pto', 'input': {'employee_id': 'E1234', 'start_date': '2023-07-10', 'end_date': '2023-07-12', 'reason': 'family vacation'}}}]
INFO:__main__:Requesting tool submit_pto. Request: tooluse_ywjLWDjfSISVP91AfBAmag
{
"text": "The PTO request has been successfully submitted for the specified dates and reason."
}
Question: What is the dress code policy?
INFO:__main__:Generating text with model anthropic.claude-3-sonnet-20240229-v1:0
[{'text': 'To get information about the dress code policy, we can use the `hr_policy` tool like this:'}, {'toolUse': {'toolUseId': 'tooluse_ZGMnL4HSRfKlZ9S8wbSZoA', 'name': 'hr_policy', 'input': {'question': 'What is the dress code policy?'}}}]
INFO:__main__:Requesting tool hr_policy. Request: tooluse_ZGMnL4HSRfKlZ9S8wbSZoA
Dress Code Policy Purpose: This policy outlines the dress code requirements for employees to maintain a professional appearance and create a positive work environment. It applies to all employees, regardless of their position or department. Policy: 1.1 All employees are expected to dress in a manner appropriate for their job roles and responsibilities. Refer to the Job Descriptions for specific requirements. 1.2 Business casual attire is required for office-based employees, including collared shirts, dress slacks, skirts, and dresses. See the Business Casual Guidelines for more information. 1.3 Employees in customer-facing roles should wear company-provided uniforms or adhere to a more formal dress code. Consult the Uniform Policy for details. 1.4 Clothing with offensive or inappropriate language, graphics, or logos is strictly prohibited. Review the Inappropriate Attire Examples for guidance. 1.5 Supervisors have the authority to determine the appropriateness of an employee's attire and may send employees home to change if necessary. Refer to the Supervisor Responsibilities for more information. Compliance: Failure to comply with this dress code policy may result in disciplinary action, up to and including termination of employment. Progressive disciplinary steps are outlined in the Disciplinary Action Policy., Document 3: Social Media Policy Purpose: This policy provides guidelines for employees' use of social media platforms, both personally and professionally, to protect the company's reputation and confidential information. It applies to all employees, contractors, and temporary workers. Policy: 3.1 Employees are expected to conduct themselves professionally on social media and avoid posting content that may harm the company's reputation. Review the Social Media Best Practices for guidance. 3.2 Confidential company information, including client data, financial information, and intellectual property, should never be shared on social media. Refer to the Confidentiality Agreement for more details. 3.3 Employees should not post disparaging comments about the company, its employees, clients, or competitors. See the Anti-Harassment Policy for more information. 3.4 When posting content related to the company, employees must clearly state that their views are personal and do not necessarily reflect those of the company. Use the Social Media Disclaimer when necessary. 3.5 The company reserves the right to monitor employees' public social media activity and take appropriate action if necessary. Consult the Employee Monitoring Policy for details. Compliance: Violations of this social media policy may lead to disciplinary action, up to and including termination of employment. Employees should exercise good judgment and seek guidance from their supervisors or the HR department if unsure about the appropriateness of their social media activity. Refer to the Social Media FAQs for common questions and answers., Document 2: Vacation Leave Policy Purpose: This policy outlines the guidelines for employees to request and take vacation leave. It ensures fair and consistent application of vacation benefits across the organization. Policy: 2.1 Full-time employees are eligible for paid vacation leave based on their years of service with the company. See the Vacation Accrual Schedule for details. 2.2 Vacation leave accrues at a rate of X days per month, up to a maximum of Y days per year. Refer to the Vacation Accrual Calculator to determine your accrual rate. 2.3 Employees must submit vacation leave requests to their supervisors at least Z weeks in advance for approval. Use the Vacation Request Form to submit your request. 2.4 Vacation leave requests will be approved based on business needs and the order in which they are received. Check the Vacation Scheduling Guidelines for more information. 2.5 Unused vacation leave may be carried over to the next year, up to a maximum of W days. Consult the Vacation Carryover Policy for details. Compliance: Employees are encouraged to use their earned vacation leave to maintain a healthy work-life balance. Managers should work with their teams to ensure adequate coverage during employee vacations. Refer to the Manager's Guide to Vacation Planning for best practices.
{
"text": "All employees are expected to dress in a manner appropriate for their job roles and responsibilities. Business casual attire is required for office-based employees, including collared shirts, dress slacks, skirts, and dresses. Employees in customer-facing roles should wear company-provided uniforms or adhere to a more formal dress code. Clothing with offensive or inappropriate language, graphics, or logos is strictly prohibited. Supervisors have the authority to determine the appropriateness of an employee's attire and may send employees home to change if necessary."
}自定义调整
- 修改
get_pto_balance()和submit_pto_request()函数中的代码逻辑，集成第三方API与大家的HR系统和数据库进行集成。 - 更新
get_hr_policy()函数，使其从大家自定义的HR知识库中检索信息。 - 调整
tool_config字典，添加、删除或修改可用工具，以适应具体的HR应用场景。
结论
本教程展示了如何使用Amazon Bedrock的Converse API和知识库功能来构建一个HR智能助手应用场景。通过利用对话式AI、工具调用（函数调用）以及与HR系统和知识库的集成，大家可以创建一个强大的HR自动化系统，用于处理HR相关查询和任务。示例代码展示了如何配置工具、处理工具请求，并使用Amazon Bedrock模型生成智能响应。
更多推荐
19
0- 0
已为社区贡献13条内容
所有评论(0)