CourseNatural Language Processing: The Complete Course · Module 10: LLMs for NLP, and Responsible NLP · part 67 of 73
Part 67 · Module 10: LLMs for NLP, and Responsible NLP

Lesson 2: Pulling checked facts out of messages with an LLM

39 min read·14 Oct 2026
Module 10 · Lesson 2 · About 50 minutes · Needs: basic Python; pydantic records (Module 8 Lesson 1) and the setup cell (course-guide/02); Module 10 Lesson 1 helps but is not needed

The situation

It is Tuesday 10 November 2026, project week 10. Dev wants three fields from every message, in any of the four languages: the order number, the product and the fault. Module 8's rules filled his booking records, but rules cannot read "der Schlüssel klemmt" (the key sticks).

His example is HT48, a hand-written German test message: "Wo ist meine Bestellung TW-122960? Und ich will das LockSure D-lock umtauschen, der Schlüssel klemmt." (Where is my order TW-122960? And I want to exchange the LockSure D-lock, the key sticks.) All data in this lesson is simulated.

HT48, GERMAN, HAND-WRITTEN Wo ist meine Bestellung TW-122960? Und ich will das LockSure D-lock umtauschen, der Schlüssel klemmt. order TW-122960 product LockSure D-lock fault der Schlüssel klemmt Dev reads: the key sticks
Dev wants every message, in any language, turned into three fields (amber: the customer's free text; blue: the record your code builds).

The rest of this part opens with an account

Signing up is free and takes one email. Nothing to pay, and nothing renews.