BC Card Automates Financial Operations with Multiple AI Models, Reduces GPU Costs by 70%

BC Card showcased its 'Agentic AI Operating Platform' at Red Hat Summit 2026 in Atlanta, becoming the first domestic financial company invited as a presenter. [Photo=BC Card]

BC Card has enhanced the efficiency of its financial AI operations by connecting multiple specialized small AI models. This approach reduces costs and increases processing speed compared to the traditional method of relying on a single large language model (LLM).
On May 14, BC Card unveiled its 'Agentic AI Operating Platform,' developed in collaboration with Red Hat, at the Red Hat Summit 2026 in Atlanta. The summit is an annual conference focused on open-source, cloud, and AI technologies hosted by the global open-source company Red Hat.
The platform presented by BC Card combines several small language models (SLMs) tailored to specific tasks, allowing for simultaneous processing instead of concentrating all tasks on one LLM. Agentic AI refers to technology that enables AI to make decisions and take actions independently, beyond just providing simple answers.
This structure has improved both the cost and speed of AI operations. SLMs can be optimized for specific tasks, yielding stable results while increasing inference speed by more than three times and reducing GPU resource usage by up to 70%.
BC Card has leveraged its experience in processing over 10 billion payment transactions annually to develop an efficient AI operation method that distributes costly GPU resources effectively. This has led to faster service and lower operational costs.
The platform is already being applied in real services. BC Card operates 'Eat.pl,' which analyzes real-time consumption patterns to recommend restaurants; 'BCGPT,' a generative AI platform that supports employee tasks; and 'MOAI,' which connects multiple AIs to automate tasks. Additionally, BC Card is making its self-developed AI models and specialized financial datasets available to the public. Approximately 38 LLMs have been released on the global AI open-source platform Hugging Face, achieving an average of over 100,000 downloads per month.

* This article has been translated by AI.

SEOYOUNG LEE 2s0@ajunews.com