I-Corps: A Framework for Streamlining the Development and Deployment of Generative Artificial Intelligence (AI) Models on Enterprise Data

University Of Texas At San Antonio, San Antonio TX

Investigators

Abstract

The broader impact and commercial potential of this I-Corps project is the development of an artificial intelligence (AI) platform to streamline and enable data scientists and researchers to deploy AI faster and more easily. Currently, the development and deployment of these models often require substantial data, time, and financial resources, which can be challenging for many organizations, particularly small and medium-sized enterprises. The proposed technology is an AI framework that integrates advanced AI technologies with practical real-world applications. It is designed to streamline the creation and deployment of large language models and is adept at building generative models based on enterprise data while ensuring stringent safeguards for data protection and generation. In addition, the proposed platform provides flexible deployment options, functioning efficiently on public cloud platforms for cost-effectiveness and scalability, or within private clouds or on-premises servers for organizations that prioritize data privacy and security. The proposed technology may be used by data scientists and researchers to simplify model development, including data acquisition, manipulation, curation, feature generation, hyperparameter tuning, and deployment, to efficiently build and deploy state-of-the-art AI modeling based on their custom datasets. This I-Corps project is based on the development of a software platform for streamlining the deployment of generative Artificial Intelligence (AI) models including multimodal large language models. The proposed technology leverages research-based data engineering and machine learning software libraries, including data-plugins, service application programming interfaces, and reinforcement learning with human feedback and parameter optimization to simplify complex data procedures. In addition, reinforcement learning with human feedback and supervised fine tuning are used to enhance the performance of AI models and improve the security and safeguards, to make the deployment of these models safer and more reliable. The proposed technology is designed to operate on public cloud platforms or private clouds/on-premises servers, which prioritizes data privacy and security for enterprises. A focus on safeguarding large language models aligns with societal imperatives for secure and trustworthy AI applications. The proposed platform uses a user-friendly interface and data plugins for tasks like data acquisition, annotation, hyperparameter tuning, fine-tuning trillion parameter models on a single DGX server for graphics processing unit (GPU) scheduling to conduct model training and operating deployment on cloud systems or on on-promise cyberinfrastructure. By automating these tasks and streamlining development workflow, the platform reduces technical hurdles, saves time, and cuts costs. This enables efficient data processing and broadens access to advanced AI development for various professionals and researchers. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.