Use of OpenAI

Affinda has integrated with OpenAI's GPT family of large language models to enable a seamless and immediate experience when creating new models and when adding new fields to existing models.

With our OpenAI connection, Affinda now offers "one-shot learning," meaning that you only need to provide one annotated example document to start making accurate predictions on subsequent documents.

This means you can begin utilizing the models almost instantly, and their performance will only improve over time. As we transition from OpenAI-generated predictions to Affinda's standard model framework, which delivers faster and more precise results when ample data is available, the quality of results will continue to excel.

When OpenAI is used

Affinda may leverage OpenAI's technology in the following scenarios:

  1. Creating a new custom extractor.
  2. Adding a new custom field to one of Affinda's standard extractors, such as invoices.


No data is shared with OpenAI when using our 'off-the-shelf' models

We do not share any data with OpenAI for model training if you are using our standard models (e.g. resume, job description, passports, etc.), even if you have given consent

How it works

  1. A new extractor or add a new custom field is created
  2. User uploads and labels a single document as an example
  3. Additional documents are uploaded for labeling and validation
  4. Affinda utilizes OpenAI to predict data labels for these documents, using the information from the first document as a 'prompt' for the predictions
  5. Once a sufficient number of documents have been labeled (typically around 50 documents), the custom extractor or field will be seamlessly integrated into Affinda's standard model framework, eliminating the need for OpenAI.

Respecting your consent

As Affinda utilizes OpenAI's technology, we value your consent to utilize this integration. Customers have the option to opt-out when creating new Collections or through the Collection Settings page at any time if they prefer Affinda not to send documents to OpenAI.


Opting out of using OpenAI does not restrict the creation of custom extractors or fields

However, it does require customers to manually label all data in the initial set of documents (approximately 50) before an Affinda model can be created without any predictions provided. This may increase the manual effort required