Rules Still Win

A team spends a week building a model to find order numbers. A simple pattern would have taken a minute. If something has a fixed format, you don't need AI. It's a good habit to ask this first.

• A week on a model
• A pattern takes a minute
• Fixed formats don't need AI

Extraction by pattern

Order numbers, dates, money amounts, and email addresses all have fixed shapes. A small dictionary of patterns finds all of them. Run it. Four clean results, from one sentence, in microseconds, and for free. It never makes things up. These results can be passed to a language model as trusted facts, or be used to check what a model extracted. Rules and models work best together.

import re

PATTERNS = {
    "order": r"\b[A-Z]{2}\d{4}\b",
    "date": r"\b(?:\d{1,2} (?:Jan|Feb|Mar|Apr|May|Jun|Jul|Aug|Sep|Oct|Nov|Dec))\b",
    "amount": r"[$£€]\d+(?:\.\d{2})?",
    "email": r"[\w.]+@[\w.]+\.\w+",
}

text = "Order AB1234 from 3 Mar cost $18.50, please email sam@example.com"
for name, pattern in PATTERNS.items():
    print(f"{name:<7} {re.findall(pattern, text)}")

order   ['AB1234']
date    ['3 Mar']
amount  ['$18.50']
email   ['sam@example.com']

Summary