Finetuning of text-generation models: Ramp strives to keep privacy in mind by training its models with anonymized versions of customer data. Ramp uses a method of first removing identifiable information (e.g., name, email, phone numbers, etc.) and replacing the data with randomly generated strings. These could be used for services like memo suggestions for receipts, and auto-matching receipts to transactions.

In-context learning (ICL) models: Ramp uses in-context learning with our models to perform inferences to service customers by referencing their existing data. The model does not retain customer data after use and customers cannot access other customers’ data or outputs. These are used to suggest merchant codes or transaction memos based on a customer’s historical patterns.

Ramp Spend Management Solution