Intermediate project

Document AI pipeline

Builds on 1 lessonabout 35 min of reading first

In this project

What you need first

Multimodal GenAI and document AIMultimodal and document AI

Skills you will practise

Document and multimodal AI

Skip to the tasks

Problem

Compare ways of extracting data from PDFs and pick a pipeline using measured accuracy.

Requirements

20 PDFs that contain tables.
A comparison of text extraction, OCR or layout parsing, and a vision-language model.
Field-level accuracy for each method.
Page citations.

Prerequisites

Multimodal GenAI and document AI

Tasks

To do: Collect 20 PDFs with tables.
To do: Extract with three methods.
To do: Measure field accuracy for each.
To do: Add page citations to the output.
To do: Choose a pipeline and say why.

Evaluation

Record real measurements and never invent numbers. The finished project should give you: parsing-quality insight and table handling.