🏦Revolut Publishes PRAGMA: In-House Transformer Model Trained on 26 Million Users' Banking Data
Revolut Unveils PRAGMA Banking AI Model
TL;DR
Revolut publishes PRAGMA, a custom Transformer model trained on 26M users and 24B banking events, powering its UK conversational assistant.
Revolut published details of PRAGMA, an in-house Transformer foundation model trained on data from 26 million users and 24 billion banking events. The model now powers Revolut's AIR conversational assistant rolling out in the UK, marking one of the first purpose-built foundation models by a digital bank.
Key Points
Custom Transformer model built in-house
Trained on 26 million users' data and 24 billion events
Powers AIR conversational assistant in the UK
One of first purpose-built banking foundation models
Why It Matters
A fintech building its own foundation model signals that major financial institutions may increasingly train domain-specific AI rather than relying solely on general-purpose models.
Frequently Asked Questions
Why does this matter?
A fintech building its own foundation model signals that major financial institutions may increasingly train domain-specific AI rather than relying solely on general-purpose models.
What happened?
Revolut publishes PRAGMA, a custom Transformer model trained on 26M users and 24B banking events, powering its UK conversational assistant.
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 3,131 builders reading daily.