用aws搭建一个基于ai的自动邮件回复系统

这是一个基于aws的ai自动回复email的尝试。用户验证邮箱之后，aws ses得到授权可以代替用户接收&发送邮件。aws ses在接收邮件的时候触发lambda。lambda提取邮件的sender，topic和content。调取open ai的api生成回复。然后lambda调取ses发送邮件给sender。

产品需求

functional requirements

这个系统应该实现以下功能：

1. 识别incoming email的意图，过滤掉广告和营销邮件

2. 收到客户发来的问询邮件，这类如果有明确的指示直接回复

3. 收到没有明确意图识别的邮件，转发给我的个人邮箱做判断。

performance requirements

1. 系统可以在几秒钟内做出自动判断。

2. 价格应该控制在1美元之内。

aws ses接收邮件的设置

1. Go to AWS SES in your Region. In my case, it is us-west-2.

2. Under Configuration -> Email Receiving, create a rule set.

3. When creating a rule set, add Recipient condition as your (business) email address. This email address will receive and automatically reply messages using openAI.

4. Next, in the Add actions section, add Publish to Amazon SNS topic. Create a SNS topic that listens to email receive events. Use Base64 encoding.

5. Go to AWS SNS. Select the SNS topic you just created. Create susbcription for your SNS topic. Select AWS Lambda for the protocol.

6. Go to AWS Lambda. Make sure your newly created Lambda listens to the newly created SNS topic. Also, make sure the SNS topic has permission to invoke Lambda.

lambda处理incoming邮件

Write a simple code snippet that extracts incoming email content.

def lambda_handler(event, context):
   print("Received the following message.")
   if 'Records' in event:
      # this is a SNS message
      for record in event['Records']:
         sns_body = record['Sns']
         message_body = json.loads(sns_body['Message'])
         encoded_email_content = message_body['content']
         decoded_email_content = decode_base64(encoded_email_content)
         plain_text = read_text_plain(decoded_email_content)
         print(plain_text)
         email_subject = message_body['mail']['commonHeaders']['subject']
         print(email_subject)
         email_sender = message_body['mail']['source']
         print(email_sender)

open ai的设计和整合

首先需要一个open ai的key，这个怎么获取不赘述了。

在aws lambda里加入open ai的layer，并且把lambda中获取的email content和subject feed到open ai里面做邮件意图识别，用open ai生成回复邮件。

我在这里省略了openai的prompt等信息。将来我会把github发在这里。

result = classify_and_reply_email(plain_text, email_subject)
   if not result['reply_needed']:
       print("Do not reply.")
       return

lambda发送回复邮件

def send_email(sender, recipient, subject, body_text, body_html):
    # Create a new SES client
    ses_client = boto3.client('ses', region_name="us-west-2")

    # Try to send the email
    try:
        # Send the email using SES
        response = ses_client.send_email(
            Source=sender,
            Destination={
                'ToAddresses': [
                    recipient,
                ],
            },
            Message={
                'Subject': {
                    'Data': subject,
                    'Charset': 'UTF-8'
                },
                'Body': {
                    'Text': {
                        'Data': body_text,
                        'Charset': 'UTF-8'
                    },
                    'Html': {
                        'Data': body_html,
                        'Charset': 'UTF-8'
                    }
                }
            }
        )
    except ClientError as e:
        print(f"Error: {e.response['Error']['Message']}")
        return False
    except NoCredentialsError:
        print("AWS credentials not available.")
        return False
    
    print(f"Email sent! Message ID: {response['MessageId']}")
    return True