New platform gives Funen companies secure access to work with Generative AI

Today, many platforms allow companies to experiment with Generative AI. The challenge is that many platforms are neither scalable, secure, nor GDPR compliant. As a result, many companies are hesitant to use the new technology—the risk is simply too high in relation to the cost and knowledge level.

In collaboration with Context&, Erhvervshus Fyn has built their own platform that allows Funen companies to use their own confidential data on a platform that is secure, GDPR-compliant, and otherwise scalable, and where the content can be lifted into the company's environment at any time when they outgrow the platform.

The platform is built on a RAG architecture. RAG stands for Retrieval-Augmented Generation and combines information retrieval with text generation. This type of architecture is often used to answer complex queries by drawing on a large amount of external knowledge and generating precise, contextual answers.

The platform is built on Azure AI Services. Although we use the GPT models as a starting point, the solution can work with alternative language models, so you are not locked into a specific language model. "It is very much the participating companies' work with their knowledge data that can be used in the model that is an important benefit for them and which they will also be able to use in a completely different Generative AI platform," says Erik David Johnson, Chief AI Officer at Context&.

The Funen companies upload their texts and documents to the platform. The knowledge they want the AI to work with is stored as a large dataset for the information search. Once the relevant documents are found, they are used as context for a language model, such as GPT, which generates answers based on the information found. This way, the retrieval process's results are combined with the model's generation capabilities to create an informed and relevant response.

The platform meets the requirements for GDPR regulations and has strong security. Uploading files and other external access is done via HTTPS and two-factor login—possibly MitID. No outsiders can gain access, and not even Erhvervshus Fyn can see the texts uploaded by companies. The chatbot is easy to add to your website, intranet, or wherever you want to use it, and it can be targeted for internal use or aimed at the company's customers.

The platform was initially developed in an MVP stage, which can give participating companies a unique opportunity to familiarize themselves with the technology and start securing their future innovation potential and competitiveness. The complete platform is expected to be ready during the first quarter of 2025.