Custom document

Extracting metadata and line items from any custom document with examples in Python, Node, Java, and C#

This is a general guide that covers how to extract data from any pseudo-structured documents with examples in Python, Node, Java, and C#. You will learn how to easily train and extract data from various different documents in many languages and character sets.

You will need:

  • Two different examples of a document
  • Correct values for at least one document
  • 15 minutes of your time

This guide shows you how to

  1. Create custom document type
  2. Add multiple suppliers
  3. Execute training
  4. Extract data from documents
  5. Continuously improve models after extraction

Getting your API Key

The Authorization header for your API key is: <<apiKey>> (Login if you do not see one).
You can also obtain the API key by visiting the Settings page.

Getting your API key

1. Create a new document type

Before you start extracting data, you need to define a document type. Navigate to the Dashboard page and click on the New document type button in the top right corner of the table. Next, select the Custom document card.

Document type is used for all your suppliers
Click on Document type to learn more.

You will have to define all metadata fields and line item fields you want to extract. The only exception is supplier_name which must be present on each document type.

To ensure consistent training and data extraction typless uses 3 field data types:

Field typeWhat is it used for?
STRINGGeneral string fields like document numbers, address, company names, payment references, IBANs, ...
DATEDates like issue date, pay due, date of service, delivery date, contract date, ...
NUMBERNumbers you want to perform calculations with like total amount, net amount, ...

Want to learn more about defining fields?
Check out the fields or line items guide to learn more.

2. Add suppliers

Once your document type is created, you need to add data to the dataset of your document type.

To add document to the dataset use the add-document endpoint or use training room where you can easily upload a file and fill out necessary infromation.

Dataset is created by uploading an original file with the correct value for each field defined inside the document type:

As you can see, to achieve high accuracy typless only needs the values that are on the document. However, there are some rules to keep in mind when providing values.

Want to learn more about providing training values?
Check out the fields or line items guide to learn more.

3. Execute training


Training is executed automatically every day at 10 PM CET

For all of your suppliers with new documents in the dataset of all your document types.
Free of charge

To immediately see the results you can trigger the training process on the Dashboard page.
Look for your document type in the list, and click on the .

Need more information about training? You can read more about it here.

4. Extract data from documents

After the training is finished, you can start precisely extracting data from documents from trained suppliers.

To add document to dataset use the extract-data endpoint.

Need more in depth explanation of the response?
You can read about it here.

5. Continuously improve models

typless embraces the fact that the world is changing all the time.
That's why you can improve models on the fly by providing correct data after extraction.
Let's say your company has a new partner Best supplier. You don't need to start over with building the dataset. You can simply extract and send the correct data after they are verified by your users.
You can learn more about providing feedback on the building dataset page.


Closed workflow loop - improve models live!

Use every action from your users to adapt and improve typless models without any extra costs.

To send feedback use the add-document-feedback with object_id.

Running typless live

The only thing that you need to do to automate your manual data entry is to integrate those simple API calls into your system.

Have any questions or you need some help? Contact us in the Messenger.