End user happiness: Alfred helps to design high-quality and conformant software architectures

GENIUS

Coding assistants are among the most successful applications of artificial intelligence to software development. However, a developer typically only spends 40% of their time writing code with a large portion of their time also working on component decomposition, performance and reliability, i.e. designing the so-called system architecture. This activity typically involves brainstorming in front of a white board, doing diagrammatic design and analysing and trading off alternative solutions. . During this process, software architects must balance established design principles with organisational constraints, such as policies governing user authentication and secure data storage.

As part of the ITEA project GENIUS, BT Group (United Kingdom) is developing Alfred, an AI-powered software architect that can assist with brainstorming, white-boarding and ensuring high-quality and conformant architecture design. Its mission is to accelerate high quality software delivery by driving towards best practice and conformance and giving easy access to the relevant information to do so.

Large language models are currently limited in their ability to do architecture design, as they lack access to company specific guidelines and are not well suited for understanding diagrams and predicting quality attributes such as performance or security. Alfred uses advanced technologies to reason about the strengths and weaknesses of architectural designs, and which best practice and company guidelines need to be considered. It achieves this by capturing the semantics and intent of this knowledge, for example the motivation for requiring encryption as the system is handling sensitive data. Developers can access Alfred using natural language or chat and collaborate on a digital white board in a way that closely mimics how human architects collaborate.

Imagine developers designing a web shop that they would like to run in the cloud. They can explain the details to Alfred, and the AI assistant will provide a diagram based on best practice and company-specific information. Alfred will ensure the design conforms to expectations and document the discussion and decisions for easy future reference. Alfred can also ask for additional information if needed, for example clarifying whether the system will be used by employees or customers or if it will handle sensitive data, and adjust the design according to the answers.

Alfred has the potential to transform architectural design into a faster and more collaborative experience. By doing so, it enables faster and higher quality designs that are better documented and easier to understand during approval and sign-off activities.

More information:
- https://itea4.org/project/genius.html
- https://genius-itea.github.io/

This development of the BT Group in the ITEA project GENIUS is supported by Innovate UK.