APC 技術ブログ


株式会社 エーピーコミュニケーションズの技術ブログです。

Dive into Building Production Quality Gen AI Applications


In this session, emphasis was placed on deeply understanding the considerations of quality and performance required when transitioning from a prototype to a product. The speakers, Miles Curtis and Eric, provided expert insights into the complexities and necessary steps involved in this transition.

Why is it difficult to build high-quality engineering applications?

Miles Curtis explained that one of the initial barriers to creating high-quality engineering applications lies especially in handling unstructured data inherent in Retrieved-Augmented Generation (RAG) applications. These applications must analyze vast datasets and generate accurate real-time responses, necessitating stringent quality control.

The Journey to High-Quality Engineering Applications

Eric emphasized that the success of RAG applications in a product environment requires continuous testing and refinement. This journey includes a series of steps, starting from proof of concept and gradually increasing in complexity. Adopting a phased implementation strategy allows for incremental improvements based on continuous feedback.

Transitioning from Proof of Concept to Product

The primary goal is to evolve an application from proof of concept to a fully functional product environment. During this progression, developers face multiple challenges including performance monitoring, debugging, and assimilating user feedback. By diligently adhering to these developmental stages, not only are user expectations met, but the delivery of a product that surpasses them is ensured.


Miles Curtis and Eric highlighted that building generative AI applications is both a technical endeavor and a methodological strategy. Prioritizing the maintenance of quality and performance, while anticipating the application’s future evolution, is crucial. The path from concept to product is iterative, demanding consistent refinement and adaptation. This deliberate approach to development ensures that generative applications are robust, efficient, and specially tailored to meet the evolving needs of users.

About the special site during DAIS

This year, we have prepared a special site to report on the session contents and the situation from the DAIS site! We plan to update the blog every day during DAIS, so please take a look.