Research on the fruitful collaboration between GenAI systems and Software Engineers to support typical software engineering tasks.


Motivation 

GenAI has the potential to improve the quality of artifacts in cooperation with experienced software engineers efficiently. Due to the generic nature of GenAI a hybrid approach should be investigated for all major artifacts (requirements, design, code, unit tests, acceptance tests). The quality focus (e.g., evolvability, green code) can be easily shifted. One major goal is the construction of a minimal prototype that supports a couple of GenAI integrated tasks:

  • Retrieving artifacts (e.g., requirements, unit tests, code) from a repository
  • Retrieving classical analysis data for these artifacts (e.g., from SonarQube)
  • Hybrid analysis of e.g., software design using LLMs and static analysis data for prompting
  • Retrieving suggestions for improvement from LLMs
  • Support for evaluation of the LLM suggestions and learning from this data

In a second work package, software evolution with LLMs is addressed. The evolution of existing (legacy) software is a major challenge. Supported by GenAI systems this task could be more manageable. Experience with a value-based migration of parts of MUSE from Perl to Python are promising. Currently, there are some method frameworks available how to evolve software – these approaches do not consider the potential of GenAI systems and therefore have to be adopted. As a goal, a method toolbox should be provided for the various tasks necessary for software evolution.

Duration 10/2024- 09/2024

Research Partner  Siemens AG

Contact Reinhold Plösch

GenAI Supported Software Engineering Tasks Experiments (AISeTa)