Speaker
Description
Calibration Operations Manager Bot for ATLAS Tile Calorimeter (COMBAT) is the next-generation of the calibration management system developed for the High-Luminosity LHC era. It combines modern AI techniques with a fully asynchronous, scalable architecture to meet the evolving operational demands of the ATLAS experiment, including the transition in the database from COOL to CREST for the conditions data management.
COMBAT provides a significant modernization aligned with current software practices, user expectations, and long-term maintainability. The system achieves high scalability through an asynchronous FastAPI-based backend and features a dual-interface design: a neural network–powered natural language interface for intuitive interaction and a structured quick-form interface, both connected to a secure backend that manages CREST operations with full validation. At its core, COMBAT employs a Multi-Layer Perceptron with TF-IDF vectorization for intent classification, enabling the AI planner to translate user queries into validated calibration actions, while providing real-time feedback, command history tracking, and data integrity assurance based on Pydantic. By introducing intuitive interfaces, guided workflows, and seamless integration with CREST, COMBAT aims to deliver a robust, scalable, and collaborative platform that supports both expert users and newcomers.
This contribution presents the design, workflow, and implementation of COMBAT, detailing the dual-interface architecture, AI-driven intent classification system, and safety validation workflows. Successful proof-of-concept deployment results are presented, demonstrating promising improvements in workflow efficiency and system usability.