Collaborative Support for Software Product Line Modeling and Project Management: A Generative AI-Enhanced Approach with Real-Time Synchronization
TL;DR
This paper presents a collaborative extension of VariaMos that combines synchronization based on Conflict-free Replicated Data Types, awareness and governance mechanisms, and integrated artificial intelligence (AI)-assisted model authoring with controlled feasibility of combining multi-artifact SPL collaboration with integrated AI-assisted authoring.
Abstract
Background: Software Product Line (SPL) engineering relies on coordinated work over variability-intensive artifacts, yet existing SPL tools provide limited support for synchronized multi-user modeling and project-level collaboration. Objective: This paper presents a collaborative extension of VariaMos that combines synchronization based on Conflict-free Replicated Data Types (CRDTs), awareness and governance mechanisms, and integrated artificial intelligence (AI)-assisted model authoring. Method: Following Wieringa’s Design Science methodology, the treatment was designed, implemented, and evaluated through unit and functional tests, collaborative proof-of-concept scenarios, controlled load experiments, an exploratory expert survey with 20 specialists, and a bounded preliminary evaluation of the AI-assisted component. Results: The environment supports project- and model-level synchronization, presence awareness, role-sensitive collaboration, comments, revision history, rollback, and vote-assisted conflict resolution. Load tests characterized the operational limits of the Yjs-based architecture and showed that partitioning users across model-specific collaborative spaces improves stability. The survey identified conflict handling, traceability, versioning, authorship, and review workflows as priorities; because it used a purposive non-probability sample, it does not support population-level or productivity claims. In a 100-case evaluation comprising 50 base prompts and 50 metamorphic follow-up prompts, a context-enhanced chatbot configuration reduced structural hallucination from 84% to 18% and increased reproducibility from 22% to 72%. These results are preliminary and limited to one LLM accessed through OpenRouter, one controlled prompt bank, and single-pass executions. Conclusions: The study demonstrates the controlled feasibility of combining multi-artifact SPL collaboration with integrated AI-assisted authoring, while identifying trade-offs among convergence, semantic control, recoverability, and latency.