CourseRetrieval-Augmented Generation: The Complete Course · Module 2: Document Ingestion and Parsing · part 15 of 95
Part 15 · Module 2: Document Ingestion and Parsing

Lesson 2: Reading PDFs, office files and email without losing answers

20 min read·13 Oct 2026
Module 2 · Lesson 2 · About 45 minutes · Needs: Lesson 1 of this module (raw files and loaders)

The situation

Kwame has a confession. In the prototype, the P9 quick-start page was typed in by hand from the PDF, which is why "how loud is the P9 on the night setting" (question q24) worked. The real manuals folder holds PDFs, some of them scans of old printed guides, Word drafts, a training deck, the parts list as CSV and spreadsheet, and partner email. Nobody will type those in.

Your job: load every format so that the answer to q24, the table row "Night, 60 m3/h, 24 dB", reaches the index intact, and so that a file you cannot read is refused with a reason, not indexed as nonsense.

THE RAW FILES KWAME HANDS OVER PDF P9 page scan P9, pixels DOCX draft PPTX training CSV, XLSX parts list email Calder Fix one loader 12 records same shape 2 rejected each with a reason
Every format goes through one loader (blue: your system). What it can read becomes records of one shape (green); what it cannot is refused with a reason (red).

The rest of this course is yours to keep

This course is bought on its own, once, and stays readable afterwards, including the parts added to it later.