Retrieval Augmented Generation (RAG) is a practical response, yet there is limited comparative evidence on how different RAG architectures trade off efficiency and answer quality for major project portfolio data. This article benchmarks two RAG pipelines using UK Government Major Projects Portfolio (GMPP) datasets (from 2020-2021 to 2023-2024) and a fixed query set, holding the foundation LLM model (i.e. GPT 4.0.), prompt template, and generation settings constant to isolate the effects of the two retrieval architectures. A vector-based hybrid pipeline (metadata filtering, dense retrieval, BM25 re-ranker) is compared with a labelled property graph-based pipeline using graph schema-aware retrieval. Results, evaluated through RAGAS metrics and a supplementary blind expert evaluation, indicate task-dependent trade-offs between the two architectures. While the vector-based RAG pipeline achieves lower query time latency and lower token consumption per query, the graph RAG pipeline improves answer quality based on RAGAS dimensions linked to trustworthiness. Particularly, graph RAG outperforms in terms of faithfulness (0.81 vs 0.78), context recall (0.87 vs 0.83) and answer relevancy (0.74 vs 0.69), suggesting better grounding in retrieved evidence and more complete coverage of the information required to provide relevant answers. Context precision is marginally higher for the vector pipeline (0.86 vs 0.85), which might be linked with the limited scoped document retrieval settings, while answer correctness performs comparably across both architectures. Expert evaluation also rated graph-based responses higher for accuracy, completeness and trustworthiness across the query subset, although preferences varied by query type. Overall, the findings highlight that architecture choice should be driven by expected query type, information risk, and the relative importance of efficiency, provenance and decision-support reliability in project and portfolio management.
The construction industry continues to experience high rates of accidents and fatalities, underscoring the need for proactive and reliable safety management. Accurate hazard identification is essential for effective image-based monitoring and decision support on construction sites. Vision–language models (VLMs) have shown strong potential for interpreting complex visual environments; however, their deployment in safety-critical applications is limited by hallucination, where hazards are inferred without sufficient evidence. To address this limitation, this study proposes a retrieval-augmented hazard detection framework that grounds VLM outputs in visual evidence through image-to-image retrieval of visually similar hazard scenarios and structured domain knowledge within a retrieval-augmented generation (RAG) pipeline. The framework integrates image-to-image retrieval of real-world hazard scenarios, supported by a curated knowledge base of 1,040 annotated construction site images, structured hazard taxonomies, and regulation-aware verification to enforce condition-constrained hazard reasoning before compliance assessment. The proposed approach is evaluated on a separate set of 260 construction site images spanning ten construction-safety hazard categories aligned with OSHA safety domains and the construction-safety literature. At the end-to-end report level, the complete pipeline achieves an F1-score of 0.917 (precision 0.939, recall 0.896) and a hallucination rate of 6.1%, substantially lower than the VLM-only baseline. At the hazard-verification gate, adding taxonomy-constrained reasoning to RAG-1 yields a precision of 0.987 and an F1-score of 0.941, while RAG-1 alone achieves an F1-score of 0.923 with 89.8% retrieval coverage. These results indicate that visual grounding provides the largest reliability gain, with structured condition verification adding further precision and gated regulatory retrieval providing traceable compliance support after hazard confirmation.
This research proposes a framework and an underpinning ontology (AccVerOnt) for the personalisation of construction safety information, instruction, and training as a means to drive a proposed construction site access verification solution. The framework incorporates established technologies, such as building information modelling and learning management systems, as a means to demonstrate a novel construction safety training solution in this regard. A methontology approach is implemented to develop the ontology. The framework as well as the underpinning ontology and the associated solution (as implemented) is evaluated by sixteen expert participants through expert interviews. A reflexive thematic analysis of the interview data is undertaken, the results of which suggest a propensity by the expert industry participants toward the solution being proposed, but an acknowledgment of the challenges associated with any attempt to implement the proposed solution at a fully personalised level at this time. Instead, having acknowledged the limitations of the delivery environment specifically in respect of hazard identification and ongoing hazard data management, as well as ongoing verification of such data against applicable regulatory frameworks, the expert participant group favoured a lesser baseline personalised version of the solution which appeared to the participants to be more commensurate with the delivery environment of the day. As a result of the findings, this research recommends a coordinated and concerted approach by relevant stakeholders to improve the conditions for the development of a mechanism by which to standardise hazard safety training content development and verification methods at the point of work site access. An end user test with twenty-two prospective end users returned only a marginal improvement in safety perception, consistent with the modest, baseline-level gains achievable in the current delivery environment.
Digital twins (DTs) are increasingly positioned as integration infrastructure for building analytics, automation, and emerging AI-enabled workflows. However, evidence remains limited regarding whether documented deployments provide the information foundations required for scalable adoption across diverse buildings and facility portfolios. This study presents a systematic review of operational digital twins in the built environment, examining semantic integration, operational connectivity, and implications for portable analytics and agent-enabled interaction. Following PRISMA guidelines, records were collected from Scopus, Web of Science, IEEE Xplore, and the ASCE Library. A total of 2,839 records were retrieved, of which 46 studies met the eligibility criteria for full analysis. Indoor environmental quality and energy were the most prevalent application areas, accounting for 34 of the 46 studies, with much of the literature addressing HVAC and building automation. Formal semantic representations were documented in roughly half of the reviewed DTs, but fewer than a third used semantic constructs to shape analytic or decision processes more directly. DTs with stronger semantic integration more often support diagnostic outputs requiring contextual interpretation across equipment and spatial relationships, whereas actions extending to direct control or automated system responses are more commonly realized through fixed, site-specific configurations. Operational outputs also remain largely confined to the platform's own interface, with limited evidence of delivery into maintenance systems, control sequences, or other governed external processes. The findings describe a field that has demonstrated meaningful operational capabilities but has not yet assembled them into coherent architectures that transfer across sites. The review indicates that machine-readable building context and lifecycle-governed metadata remain important constraints on portability, workflow integration, and the structured access needed for emerging agent-enabled interaction. Semantic infrastructure is accordingly framed as a lifecycle information requirement, one that should be established early in the building lifecycle and maintained as buildings evolve, to support durable, scalable, and AI-ready building operations.
The construction industry lags behind other industrial sectors in terms of digitalization. The construction sector's slow adoption of comprehensive digital transformation is due to significant fragmentation throughout the building lifecycle and numerous stakeholders with diverse interests and priorities, hindering effective coordination and thus delaying innovative technology implementation. Following a design science approach, we propose and develop a process-oriented analysis framework for servitization based on service-dominant logic and Gaia-X-based ecosystem standards and technologies. The servitization approach is illustrated by applying it to the building permit process in general and described with exemplary services. The novelty of the study lies in combining service-dominant logic with a process-oriented, design-science framework that is operationalised via Gaia-X ecosystem standards and technologies to systematically derive digital service opportunities along the building lifecycle.
