Revolut Scales PRAGMA Foundation Model for Real-Time Banking
Revolut has expanded its PRAGMA foundation model from research into production systems for real-time banking services, including recurring-payment detection, instant credit decisions and in-app recommendations. The model was developed with NVIDIA and trained on 26 million anonymised user histories.
At NVIDIA GTC Berlin, Revolut executives Anton Repushko and Amey Baokar described the infrastructure behind deployment. The platform pre-processes and tokenises customer activity outside the decision-making request path, and serves an initial base of 80 million customers. It is designed to scale threefold and support up to 10,000 events per customer. Revolut targets a double-digit millisecond p99 data-read time within a sub-second end-to-end inference budget.
For efficient GPU use, Revolut keeps a shared model backbone resident and swaps in task-specific LoRA adapters, which adjust 2–4% of core parameters. It also uses dynamic batching to process multiple users together. The company says immutable data pipelines, versioned tokenisers and audit trails support regulatory requirements.
Revolut pairs PRAGMA’s event-sequence modelling with STRIDER, a tabular foundation model, for cases such as applicants with limited transaction histories or external credit data. The announcement highlights the growing use of AI infrastructure in financial services, but does not report new cryptocurrency products, trading activity or performance results that would directly affect crypto markets.
Neutral
The announcement is about Revolut’s deployment of AI models for banking operations, not cryptocurrency, blockchain infrastructure or digital-asset trading. It provides no direct catalyst for crypto prices, market liquidity or regulatory conditions, so the most appropriate market view is neutral.
In the short term, traders may treat the news as a modest signal of broader financial-sector investment in AI, but it does not identify a crypto-linked company or token likely to see an immediate change in demand. Similar announcements about banks adopting machine learning have generally been more relevant to the firms’ technology strategy than to broad crypto-market direction. Any reaction would likely be limited to sentiment around financial technology and AI-related assets, rather than Bitcoin or the wider digital-asset market.
Over the longer term, more automated credit and fraud systems could influence how financial institutions manage risk and serve customers. That may indirectly affect fintech competition and the adoption of digital financial services. However, the article gives no indication that Revolut plans to connect PRAGMA or STRIDER to crypto markets. Traders should therefore look to separate market drivers—such as crypto-specific regulation, liquidity, institutional flows and macroeconomic conditions—for clearer signals on price direction and stability.