Recent advances in machine learning and scientific visualization have revolutionized the industry with novel applications and services that deeply impacted human life mainly. However, state-of-the-art machine learning models, especially deep models, are highly dependent on reliable data alongside their respective annotations. Unfortunately, label annotation is a time-consuming and costly task, especially for applications that require manual annotation performed by experts with special skills and domain knowledge. In this scenario, methods that assist experts to perform annotations effectively have a major impact on deploying reliable machine learning models. This article proposes a modern web application, called Annotat3D, that implements an interactive segmentation workflow at the Sirius/LNLS facility, enabling experts to perform fast and accurate tomographic volume labeling in high-performance computing (HPC) environment. As a result, the proposed application greatly impacts the workflow of both experts and non-experts users of Sirius’ beamlines, since most of the tools available for image analysis and visualization are not optimized to operate in HPC environments. Our methodology takes advantage of recent developments in network communication to efficiently exchange data load in main memory between two-node clusters and modern web-based frameworks that allow us to build an efficient, clean, and simple graphical user interface.