Unit Testing generation for embedded software using LLM

Unit testing is the first level of tests, where a unit or module is checked to verify the correctness of its behaviour and whether it is compliant with its requirements. A correct test suite is capable of identifying bugs and validating that the behaviour is as expected. Allowing developers to solve and prevent errors during the integration of the module into the system or in later phases. The problem is that common practices for developing test suites rely on manual effort and static analysis tools; this makes the activity time-consuming and resource-intensive. The advancements in generative artificial intelligence (AI) and the capability of large language models (LLM) to analyse code and generate it based on different specifications have made it possible to start developing code using these tools. However, These tools must be used conscientiously, not only by delegating the activity they perform but also by checking that the code they generate is adequate to verify that the implementation meets the requirements. This thesis presents a framework that, using an AI agent, MCP servers, and various prompts; generates a unit test suite for the diagnostic event manager (DEM) module, part of the battery management system of Beond, based on the software requirements presented by AUTOSAR.