An Open-source Cross-Industry and Cloud-agnostic Generative AI Platform

Generative AI (GenAI) has recently gained immense popularity within the Machine Learning (ML) field due to its potential for enabling a wide array of applications. Businesses globally are eager to leverage this transformative technology; however, they encounter several challenges in utilizing the technology, including initial and running costs, data protection, infrastructure, adaptability across use-cases, accuracy, ethical implications, and the burgeoning carbon impact of GenAI, which act as barriers to entry. A low-cost, low-carbon impact GenAI platform that exclusively uses open-source components and smaller task-specific Large Language Models (LLMs) is necessary.This paper presents a GenAI platform that employs open-source components and models orchestrated for multiple cross-industry use-cases. The platform is designed to be cloud-provider agnostic, highly secure, scalable, and requires low infrastructure to deploy, thereby using comparatively less energy and having a lower carbon impact. These features enable several advantages, including domain transference, localization in different languages using prompt tuning, and easy deployment of complex multimodal, cross-industry, cross-domain use-cases. We have tested our platform on numerous use-cases across several key patterns, including text-to-text, text-to-code, text-to-analytics and text-to-image. All these configurations provide a semantic interface to unstructured text, tabular data, images, videos, and speech. The platform, the methodology, and ideas presented in this paper will provide enterprises with a blueprint of how to orchestrate GenAI using open-source components.Our experience building a multi-modal open GenAI platform and our investigation and analyses of primary patterns in enterprise use-cases for GenAI suggest that the proposed low-cost, low-carbon, model-to-the-data platform has the potential to revolutionize the adoption of GenAI technologies within business enterprises. We believe that the proposed platform can be transformative in this space, and we anticipate its future applications with excitement.