Digit reduces AI usage costs with GPU optimization

Digitro Tecnologia has implemented a system of orchestration of language models that reduces the demand for GPUs (graphics processing units, originally created for images and today used in artificial intelligence). The innovation ensures efficiency, scalability and more competitive costs in applications ranging from digital service to advanced data analysis.

The GPU, is a specialized electronic circuit originally designed to accelerate the creation of images and videos. Its immense ability to perform large calculations in an agile way and tasks that require a large volume of data is used in resources with artificial intelligence.

According to Agenor Pacheco Junior, Product and Innovation Manager at Digitro, large language models are extremely heavy. “With orchestration, we were able to trigger only the required model at the right time, using the same infrastructure.This makes AI economically viable for diverse applications. Our goal was to tackle one of the biggest challenges of artificial intelligence in the corporate world: the cost, the, optimizing resources and delivering artificial intelligence in an accessible, efficient and scalable way“. explains the executive.

According to Agenor Junior, the great challenge overcome in this project was to bring performance savings, optimizing the use of GPUs and presenting short response times, required by real-time applications.

Innovation underpins AI services

In practice, orchestration supports AI services already embedded in the company's solutions. Among the available features are natural language processing (NLP) to interpret open questions, RAG (Retrieval-Augmented Generation), which creates a specific knowledge base of each organization, and computer vision, capable of extracting information from images, such as data from health plan cards for appointment scheduling.

The development began in October 2024 and is now in operation, available for use in all Digitro solutions, but with a particular focus on the automation of service and analysis of dialogues in contact centers, where the use of artificial intelligence is already part of the day-to-day operations