A configurable AI assistant SPL

A feature model describing a configurable AI Assistant SPL.

Along with the UVL model, we provide a complete natural language description of the SPL. This text is specifically designed to be used as a prompt for Large Language Models (LLMs) to evaluate their capability in automatically generating the corresponding feature model.

Natural language description of the model:

An AI assistant PL based on LLMs. Every valid configuration includes a Model, a Plan, and an Access option. Also, it can optionally has FineTuning, Temperature, ImageGeneration, and WebSearch. Exactly one Model must be selected among GPT4o, Gemini, or Claude, each one with a different cost and context window size represented by the attributes cost and tokens. Similarly, exactly one subscription plan must be chosen between Free and Premium, with up to 50 and 1000 queries per month, respectively. At least one access mechanism must be included: WebApp, API, or both. The API option has an additional cost of 2. FineTuning enables advanced customization at an extra cost of 5, while Temperature controls response variability and is defined as an integer feature between 0 and 5. ImageGeneration and WebSearch provide complementary functionalities. Several cross-tree constraints restrict valid configurations. Image generation is only supported when using GPT4o or Gemini. The API option cannot be combined with WebSearch. FineTuning is available if and only if the Premium plan is selected. Features are enriched with attributes such as numerical values (cost, monthLimit) and string values to model context size (tokens). The overall sum of the costs of the selected features must remain below 15.