FORTHought: Self hosted AI stack for physics labs (and more) built on OpenWebUI (github.com)

A researcher has announced the development of a self-hosted AI research platform tailored for physics labs, using Open WebUI to streamline scientific workflows such as literature reviews, spectroscopy, and data analysis. This platform is notable for its customization and optimization of existing open-source tools, facilitating accessible and reproducible research practices. By running on two machines—a Docker server for hosting services and a compute server for GPU-intensive tasks—the platform effectively integrates various functionalities like document parsing, vector search, and model reranking through a unified configuration. The platform's architecture enables the simultaneous use of multiple AI models for diverse tasks, leveraging AMD's dual R9700 GPUs for efficient processing. Key innovations include custom tool servers for automating analysis in OriginPro and advanced X-ray diffraction, as well as features for extracting and referencing academic papers systematically. With an emphasis on maximizing resource use and maintaining data privacy, the platform empowers researchers to conduct complex experiments without relying on external cloud infrastructure, marking a significant step in the intersection of AI and scientific research.