Part 87 · Module 12: Multimodal Models

Lab: One checked intake for every attachment

41 min read·13 Oct 2026

Module 12 · Lab · About 70 minutes · Needs: lessons 1 to 3 of this module, and the Module 1 and Module 6 labs (or the cells in Step 1)

What you are building

Every attachment goes through one entry point and one check against the billing record (grey): agreement is accepted (green), anything else goes to a person (amber).

It is Friday 21 August 2026, the end of week 12. Customers now attach billing screenshots and invoice PDFs, and send voice messages from the mobile app. The three lessons built a reader for each in notebook cells, under different names: check_reading, two different check_fields and three record shapes. This lab moves them into one file, supportdesk/intake.py, with one entry point, intake(kind, payload), and one record shape for every attachment. Every number read from an attachment is a claim. One function, check_record, compares it with the billing record, and any problem sets needs_human. The running example is Ana Duarte's T-1001 again, now arriving as a screenshot, a PDF, two scans and a voice message, all made in code. Three cells call the hosted model; everything else runs on your machine.

What you are building

The rest of this course is yours to keep