Generating Specifications from Requirements Documents for Smart Devices Using Large Language Models (LLMs)

The current contribution of artificial intelligence based Large Language Models (LLMs) in supporting the requirements engineering and design phase of software centered systems is analyzed. The application domain is focused on smart devices and services for Ambient Assisted Living (AAL), e.g. programmable smartwatches. In the realm of AAL, programmable smartwatches offer significant potential to support elderly users in their daily activities. However, developing applications for these devices is resource demanding, algorithmically difficult and requires careful consideration of various factors, including user-specific needs (e.g. a diminished natural sensation of thirst) and technical constraints (restricted computational power due to limited battery capacity). Our approach first utilizes widespread LLMs like ChatGPT, BARD to automatically interpret and enrich product concept catalogues, targeting at comprehensive, consistent and tailored requirements specifications for the specific domain of application. Secondly, we analyze in which extent LLMs today can contribute to the original design phase by introducing suitable functional principles and reference architectures for fulfilling the services specified as requirements including the constraints on its operations. To demonstrate the challenges and perils of this approach, we will detail the process using the specific use case of automatic drinking detection and providing suitable advice for preventing dehydration for elderly users. We will discuss and differentiate the principal strength of the approach from its actual limitations by the presently available LLMs, which are expected to increase and elaborate their generative capabilities rather frequently.