The AI4SWEng project advances AI‑powered software engineering through approaches that are efficient, trustworthy and human‑centred. A central pillar for realising the project vision is a foundational project implementation process that commits to in-depth requirements analysis and prioritisation. Rather than focusing on implementation details, this work establishes the structural and conceptual backbone that identifies and defines the scope of innovations while ensuring compliance, transparency and scalability.
At the heart of these activities lie the adoption and extension of UI‑REF [1], the User‑Intimate Requirements Hierarchy Resolution Framework. UI‑REF grounds the requirements engineering process with a structured and user‑centric methodology for eliciting, refining, and prioritising requirements with the full participation of a diverse range of stakeholders. In AI4SWEng, this methodology has been applied to support AI‑ready software engineering by embedding governance, ethics, and regulatory considerations directly into technical
requirements.
Figure 1 illustrates how UI‑REF leverages stakeholder needs to create and refine structured requirements, ontology‑based representations, and ultimately a coherent framework architecture with measurable KPIs.
A key achievement of this work is the systematic transformation of requirements into machine‑interpretable knowledge structures. Instead of treating requirements as static text rtefacts, AI4SWEng applies ontology‑based requirements engineering (OBRE) to capture both functional and non‑functional aspects in a formal, reusable, and verifiable way. This approach enables consistent interpretation of requirements across tools, teams and use cases while supporting traceability from high‑level objectives down to architectural
elements.
The motivation and principles behind introducing OBRE in AI4SWEng have been presented in an earlier AI4SWEng article, “AI4SWEng: Strong Foundations, Shared Vision, and Momentum for What Comes Next”.
To operationalise this approach, AI4SWEng uses customised GitHub Projects workspaces as shared interactive platform for managing requirements, specifications, tasks, and feedback. Requirements and use‑case specifications are defined, refined, and discussed collaboratively, with versioning, traceability, and accountability embedded by design. This setup allows stakeholder feedback, prioritisation decisions and requirement evolution to be captured transparently and linked directly to development activities.
Crucially, this requirements and specifications layer is designed to be tightly connected with the development environment itself. By linking structured requirements and architectural elements with implementation tasks, code artefacts, and validation activities, AI4SWEng establishes a fully accountable development process. Each design decision, change, or
implementation step can be traced back to its originating requirement, supporting auditability, reproducibility, and responsible AI engineering throughout the lifecycle.
Trustworthiness is therefore addressed by design, not as an afterthought. Requirements related to data provenance, privacy, security, sustainability, and regulatory alignment are treated as cross‑cutting concerns rather than isolated constraints. Integrating these dimensions early during requirements analysis, prioritisation and architecture specification, reduces downstream integration risks and lays the foundation for compliant AI‑driven
software development.
An important outcome of this work is the creation of use‑case‑specific ontologies, which contextualise general project requirements within concrete application scenarios. Figure 2 presents a high‑level view of the Medical Use Case-specific ontology, showing how domain context, functional and non‑functional requirements, and knowledge graphs are connected to AI4SWEng tools. Theseontologies act as semantic bridges between human understanding and automated reasoning, enabling traceability, validation, and future automation of development and verification processes.
Beyond technical artefacts, this foundational effort introduces collaborative, transparent, and tool‑supported requirements management practices aligned with modern DevSecOps workflows.
In summary, AI4SWEng establishes a governance‑driven, ontology‑based approach to requirements analysis and framework architecture specification, enabling trustworthy AI engineering, reducing integration risks, and creating the conditions for scalable, explainable, and fully accountable AI‑powered software development.
References
[1] Badii, A., & Fuschi, D. L. (2008). User‑Intimate Requirements Hierarchy Resolution Framework (UI‑REF) in work‑flow design for 3D media production & distribution. proceedings e report, 171.
