Generative AI-Powered Assistant for Pattern Recognition in TJ-II Stellarator

A secure generative artificial intelligence (AI) platform with nuclear fusion-specific data has been implemented for TJ-II stellarator. This includes data migration, integration, and the study of the expected feedback using chatbots, especially based on IBM Watson Assistant. The creation of the hybrid cloud and its connectivity, as well as data integration and the use of a vector database (DB), was accomplished. Subsequently, and in parallel with the processed data from TJ-II, specialized large language models (LLMs) were tested and validated to connect them with other machine learning (ML) models that were previously generated by complex algorithms and processes commonly used in the TJ-II device. From here, the ultimate goal was to develop relevant use cases related to pattern recognition and image classification issues within signals and spectrograms. It was addressed using generative AI services and reinforcement learning of the previously generated LLM models. In addition, this task aims to use this platform as a common framework for extracting significant and hidden insights during the TJ-II operation that can be scientifically validated in a conversational way.