[1/2] Deductive Software Architecture Recovery via Chain-of-thought Prompting (Category Architecture)

I read it before bed yesterday. fancy whitepaper SAR process (Software Architecture Recovery) using LLMs. I thought it was interesting to talk briefly about it.

1. The standard approach to Software Architecture Recovery usually worked bottom-up. (bottom-up). Conventionally, some code analyzer started building architectural metrics that somehow characterized the architecture of the project. (Software Architecture Metrics: Case Studies to Improve the Quality of Your Architecture told)
2. In this paper, the researchers decided to go top-down. (top-down) and start with a reference architecture, and then classify parts of the code as belonging to a particular part of that reference architecture. This allows you not only to collect the current state of the architecture as it was in the standard approach, but also to assess the discrepancy between the reference architecture of the project and what we actually have:)
3. The authors say that this approach is more similar to how software engineers work, since engineers usually know what the basic architecture in a project is, so they can use this knowledge when studying code.

Further description of the process in next postThe main illustrations with a description of the process, a description of the indicators of the components of the architecture, as well as a prompt for the LLM, which is used in the classification.

#Architecture #Software #Metrics #LLM #AI #ML #Engineering #RnD