The architecture of F2 AI Assistant
F2 AI Assistant is a fully integrated module in F2, which adheres to F2’s existing access restrictions. This page describes the overall architecture of F2 AI Assistant with a focus on its individual components and the data exchange between them.
| Contact cBrain for further information. |
The following sections cover the individual components and the interactions between them.
Components of F2 AI Assistant’s architecture
F2 AI Assistant’s architecture is made up of five main components. The relationship between them is shown in the figure. Each component is described below.
F2 database
F2’s standard architecture consists of an SQL database that stores all data from F2. This includes record, case, and document metadata as well as process data and logs. The database can be hosted on-premises or in the cloud. The following description assumes an on-premises solution where data do not leave the local setup.
F2 client
The F2 client contains F2’s standard business logics to support a secure governance model and process the relevant data. It is here, in the application layer, that users through a user interface can access and modify data based on the access control defined by the business logic.
Large Language Model server
cBrain hosts an on-premises Large Language Model (LLM) server. The server has a foundation model from cBrain installed, which can be used as the LLM engine for F2’s AI solutions. This model can be continuously updated when newer and better LLMs are developed.
It is also possible to deploy your own LLM or use a third-party provider.
Natural Language Processing user interface
cBrain has developed several user interfaces for Natural Language Processing (NLP), which provide access to an LLM. The user interfaces let you ask questions which the LLM can answer based on your appended input.
F2 AI Assistant is an example of an application of the user interface layer embedded directly in F2’s user interface.
Vector database
A vector database stores information as vectors. Simply put, vectors are numerical representations that can be used to represent various types of data objects. The vector database can effectively index and search through large sets of unstructured and semi-structured data. This is a prerequisite for the LLM to access the data input.
The vector database is installed in the same location as the F2 database. It is therefore protected by the same security mechanisms as the F2 database, and both databases are continuously synchronised as documents are stored in F2.
F2 AI components interaction
Jointly, the five components enable the use of an integrated AI solution such as F2 AI Assistant. The figure below illustrates how the components are connected. The numbers on the figure refer to the descriptions in the table that follows.
| Number | Description |
|---|---|
1 |
Data exchange between F2’s database and client All data are kept in the F2 environment. Nothing leaves the closed on-premises setup. |
2 |
Vectorisation of documents Before the LLM can access the document used for context, the document’s contents must be transferred to a vector database. This transfer happens every time a document is saved in the F2 database, ensuring that all records and documents are accessible and up-to-date. |
3 |
NLP user interface for the dataset F2 LLM (or another LLM) is used to display a chat interface that allows you to ask questions about the dataset. You can ask general questions about the dataset and receive replies based on the selected function. Access to the specific data elements is subject to F2’s access control. |
4 |
Queries from the F2 client to the LLM server F2 AI Assistant provides users with a number of standard prompts. In order to use these prompts, you must first select a chat, record, or document to serve as context. Your prompt and the context is sent as input to the LLM. Data control is maintained as users can only find and send data to which they already have access. This means that users cannot request a summary of data if they don’t have access to it already. |