02. Email Processing Pipeline

An email processing pipeline retrieves emails, prepares them for analysis, runs AI inference, and takes action based on results. The pipeline pattern appears in every automation chapter because the components transfer across use cases.

The basic pipeline structure:

import imaplib
import email
from email.header import decode_header
import json
from ollama import chat

def fetch_unread_emails(mail_server, username, password, mailbox='INBOX'):
    """Connect to IMAP and fetch unread emails."""
    mail = imaplib.IMAP4_SSL(mail_server)
    mail.login(username, password)
    mail.select(mailbox)
    status, messages = mail.search(None, 'UNSEEN')
    email_ids = messages[0].split()
    
    emails = []
    for email_id in email_ids:
        status, msg_data = mail.fetch(email_id, '(RFC822)')
        raw_email = msg_data[0][1]
        msg = email.message_from_bytes(raw_email)
        emails.append({
            'id': email_id,
            'from': msg['From'],
            'subject': msg['Subject'],
            'body': get_email_body(msg)
        })
    
    mail.logout()
    return emails

def get_email_body(msg):
    """Extract text body from email message."""
    body = None
    if msg.is_multipart():
        for part in msg.walk():
            content_type = part.get_content_type()
            if content_type == 'text/plain':
                try:
                    body = part.get_payload(decode=True).decode()
                except:
                    body = str(part.get_payload())
                break
    else:
        body = msg.get_payload(decode=True)
    return body if body else ""

This script connects to any IMAP-compatible email server, retrieves unread messages, and extracts their content. Gmail, Outlook, Exchange, and most email providers support IMAP.

The next step is choosing a model for processing. Smaller models like llama3.2:1b or 3b handle high-volume classification tasks where speed matters. Larger models like llama3.1:8b handle nuanced tasks requiring better comprehension. The choice depends on your hardware and the complexity of the required understanding.

Local verification checkpoint

Run the smallest example from this chapter in a local workspace and record the package version, runtime, data path, and observed output. If the result depends on model size, vector count, CPU/GPU backend, or available memory, note that constraint beside the exercise so the lesson remains reproducible.

Local verification checkpoint

Run the smallest example from this chapter in a local workspace and record the package version, runtime, data path, and observed output. If the result depends on model size, vector count, CPU/GPU backend, or available memory, note that constraint beside the exercise so the lesson remains reproducible.