Case Study: Ramp achieves faster, more accurate receipt processing with Modal

How Ramp automated receipt processing with fine-tuned LLMs

Ramp, a company that provides a unified platform for corporate spend management, faced challenges with fine-tuning large language models (LLMs) for their automated receipt processing. They needed greater customizability, security, and cost-efficiency than what was available from other LLM providers, requiring a solution that offered more control over their fine-tuning workflow. They turned to vendor Modal for its platform to run Python functions in the cloud.

Using Modal, Ramp built a custom experimentation framework to train many models in parallel, persist the results, and serve inference endpoints. This solution allowed them to accelerate development and significantly improve their text-to-JSON model for receipts. The results were a 34% reduction in receipts needing manual review and an estimated 79% reduction in infrastructure costs compared to other major providers. Modal also enabled Ramp to speed up other batch processing tasks, such as stripping PII from thousands of invoices in minutes instead of days.

Ramp

Gennady Korotkevich

Software Engineer