Software Inconsistency Resolver

From Computer Laboratory Group Design Projects
Revision as of 08:18, 5 November 2024 by afb21 (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigationJump to search

Proposed by Jasmin Jahić <jj542@cam.ac.uk>, to be discussed with Chris Cooper-Bland of Endava for 2026

This is currently phrased as a research project, so a specific problem and technical approach would need to be defined.

Your task is to make a architecture design consistency tool, probably based on LLM technology, that works on the different levels of abstractions. In its basic instance, the tool works with a model of software architecture and source code. It identifies parts of code belonging to abstract architectural components, suggests issues with breaking isolation, updates all the identifier names in a code base, for internal consistency and to ensure agreement with software architectural model (interfaces, component names, identifiers, other specifications). To create a software architectural model, we will use draw.io as the tool, as it enables generating clean xml from a visual representation. Besides the main goal of the project, the students will have the opportunity to explore:

- Can LLM read binary on any level and extract any software architecture-relevant information?

- Can LLMs enable control of source code from architecture (e.g., "notice" any change in architecture and suggest changes in the source code)?

- Can LLMs enable updates to architecture model based on the source code (back propagate any change from source code and conclude if anything should be changed in the architecture)?