This section provides a curated and annotated set of high-signal references drawn directly from the sources identified during the development of the Hub. Each entry below is organised under a thematic heading, presented with the link to the original source, and accompanied by substantive commentary derived from the content of each resource. This is not a passive bibliography. It is an active reading guide that explains what each resource contains, why it matters in the context of GenAI and construction, and how practitioners should engage with it. All links were verified and accessed during the preparation of this document.
The resources are organised into three thematic groups: information management, BIM, and open standards; responsible AI, risk, and governance; and public sector procurement context for the UK. A fourth group addresses the technical infrastructure of GenAI platforms, covering the major cloud platforms through which language model capabilities are accessed and deployed in enterprise construction contexts.
The resources in this group address the foundational information management frameworks, open data standards, and classification systems that underpin responsible GenAI deployment in construction. A GenAI system is only as good as the information it draws on, and these standards define what good information looks like: structured, versioned, correctly attributed, and machine-readable. Practitioners and technical architects should treat these resources as prerequisites rather than optional background, because the quality gate between information management practice and GenAI output quality is direct and unforgiving.
https://www.iso.org/standard/68078.html
ISO 19650-1:2018 is the international standard that defines the concepts and principles for information management at the level of maturity described as building information modelling across the whole life cycle of built assets. It was published in December 2018 by ISO Technical Committee TC 59/SC 13 and remains the definitive international reference for how information is exchanged, recorded, versioned, and organised across the built environment. The standard covers strategic planning, design and engineering, construction, day-to-day operation, maintenance, refurbishment, and end-of-life, making it applicable at every stage of the asset life cycle rather than only at the design and construction phases where BIM is most commonly discussed.
The standard is relevant to GenAI deployment because it defines the information quality and management principles that must be in place before project documents can reliably feed AI knowledge bases. The concepts of a single source of truth, status-based information release, version control, and clearly defined information containers map directly onto the requirements of well-functioning RAG systems. An organisation that has not yet implemented ISO 19650-aligned information management will typically find that its document estate is too inconsistent, too poorly versioned, and too lacking in reliable metadata to support GenAI workflows that produce trustworthy outputs.
As of June 2026, the standard has been confirmed as current following a systematic review completed in 2024, though a draft revision, ISO/DIS 19650-1, is under development and expected to replace the current edition within the coming months. Practitioners planning long-term Hub deployments should monitor the progress of the draft revision through the ISO website, as changes to the conceptual framework in the revised version may affect how information management requirements are expressed in Hub guidance. The standard is available for purchase from the ISO store at CHF 179 in PDF and ePub formats.
For UK construction organisations, ISO 19650 is applied through the UK BIM Framework, which provides national annexes and supplementary guidance developed by the BSI and the UK BIM Alliance. The UK BIM Framework materials remain freely available online and provide a more accessible entry point into ISO 19650 concepts than the standard itself for practitioners without a background in formal standards.
This guidance document is part of the UK BIM Framework series, which provides nationally specific guidance to support the application of ISO 19650 in UK construction projects. Part D focuses on developing information requirements: the process by which clients and appointing parties define what information they need, when they need it, and in what form, before a project begins. This process is fundamental to effective information delivery and has direct relevance to GenAI deployment because the quality of information requirements definition upstream determines the fitness for purpose of the documents that downstream GenAI systems will process.
The document explains the distinction between organisational information requirements, asset information requirements, and project information requirements, and describes how these different levels of requirement are connected in a coherent information strategy. It covers the development of the employer's information requirements document, which sets out what the client expects to receive, and the development of the BIM execution plan, which describes how the supply chain will meet those requirements. Both documents are relevant to GenAI contexts because they establish the information governance framework within which AI-assisted document processing and synthesis will occur.
For practitioners building or configuring Hub knowledge bases from project documents, Part D provides the conceptual language needed to understand why certain documents are authoritative and others are not, why version status matters, and how information containers relate to each other within a structured project information model. This understanding is essential for making good decisions about document ingestion: which documents should be included in a knowledge base, at what status threshold, and with what metadata. Practitioners using the Hub who have not previously engaged with ISO 19650 are recommended to read Part D before attempting to configure a project-specific knowledge base.
https://technical.buildingsmart.org/standards/ifc/
The Industry Foundation Classes, or IFC, is the primary open international standard for the digital description of the built environment. Maintained by buildingSMART International and published as ISO 16739-1:2024, IFC is a vendor-neutral, open schema that codifies the identity, characteristics, and relationships of objects, processes, and people involved in the design, construction, and operation of built assets. It covers physical components such as structural elements, mechanical and electrical systems, and manufactured products, as well as abstract concepts including cost breakdowns, work schedules, and performance assessments.
The IFC schema is the technical deliverable through which buildingSMART fulfils its mission of promoting openBIM, a working methodology based on open standards for the exchange of digital information throughout the built asset life cycle. Software vendors across the building information modelling ecosystem implement IFC export and import interfaces in their products, enabling interoperability between tools from different vendors. Since 1997, IFC has been tested across many projects globally and is recognised as the default format for model-based information exchange in BIM-enabled construction workflows. The schema can be encoded in multiple formats including XML, JSON, and STEP, and can be transmitted via web services or managed in linked databases.
For GenAI in construction, IFC is relevant in two ways. First, as the standard exchange format for geometric and semantic model data, IFC files are increasingly being explored as a source of structured information for AI query interfaces that allow practitioners to interrogate building model data using natural language. Second, the attribute structures and classification systems embedded in the IFC schema provide a reference point for the metadata frameworks used in construction AI knowledge bases, ensuring that asset data captured during design and construction is structured in a way that supports downstream AI processing in operations and facilities management contexts.
The buildingSMART technical portal also provides access to the IFC specifications database, which contains the detailed schema specifications for each IFC release, as well as Model View Definitions that define subsets of the schema relevant to specific exchange scenarios. Practitioners developing technical integrations between construction platforms and Hub knowledge bases will find the IFC implementation guidance and sample files available through the portal useful for understanding how IFC data is structured and how it can be processed programmatically.
https://technical.buildingsmart.org/standards/ifc/ifc-schema-specifications/
The IFC schema specifications database provides access to the formal technical specifications for each version of the IFC standard. This includes the complete entity and property set definitions for IFC 4.3, which is the current internationally harmonised version published as ISO 16739-1:2024, as well as earlier versions including IFC 4.0 and IFC 2x3, which remain widely implemented in the software tools currently deployed in UK construction practice. The database is the authoritative technical reference for developers implementing IFC interfaces and for practitioners who need to understand how specific building elements and their properties are represented in IFC-compliant data.
For the Hub's purposes, the specifications database is most relevant to the data engineering and GenAI architect roles described in Section 10. These roles need to understand the structure of IFC data to design effective pipelines for processing model-derived information in GenAI workflows, and to configure metadata schemas that align with IFC attribute structures. The database also supports the information management lead in understanding how IFC property sets relate to the asset information requirements defined under ISO 19650, enabling the development of integrated information management frameworks that span both BIM coordination and AI knowledge base population.
https://www.buildingsmart.org/standards/
The buildingSMART International standards page provides an overview of the full suite of openBIM standards maintained by the organisation, extending beyond IFC to cover the workflow and process standards that govern how information is specified, delivered, and validated in openBIM projects. Key standards in this suite include the Information Delivery Specification, or IDS, which provides a machine-readable format for defining and validating information requirements; the BIM Collaboration Format, or BCF, which enables issue tracking and coordination across model authoring platforms; and the Information Delivery Manual, or IDM, which describes the business processes and information flow requirements for specific BIM use cases.
The IDS standard is of particular relevance to GenAI in construction because it enables the automated validation of information quality against defined requirements, which is an essential precursor to reliable GenAI knowledge base population. An organisation that uses IDS to define its information requirements can, in principle, implement automated checking pipelines that validate documents against those requirements before ingestion into a knowledge base, ensuring that only conformant information enters AI workflows. This is a more rigorous approach to information quality assurance than manual checking and is consistent with the data engineering principles described in Section 10 of this document.
The BCF standard is relevant to agentic AI workflows in construction coordination contexts, where AI systems that identify design issues or coordination clashes need to generate structured issue records that can be communicated across the project team. BCF provides a standardised format for these records, enabling AI-generated issues to be integrated directly into existing coordination workflows rather than requiring manual reformatting. The buildingSMART standards page provides links to the specifications, guidance, and software certification information for each standard in the suite.
https://www.thenbs.com/our-tools/nbs-national-bim-library
The NBS National BIM Library is the UK's primary free resource for BIM object content, providing architects, engineers, and construction professionals with manufacturer-specific and generic BIM objects that can be incorporated into design models. It is maintained by NBS, which is part of Byggfakta Group, and provides access to thousands of objects representing building products and systems in a range of formats compatible with the major BIM authoring tools used in UK practice. Objects in the library are developed to a consistent standard of information quality and include the structured property sets that are required for downstream asset information management and operations purposes.
For the Hub, the NBS National BIM Library is relevant as an example of how structured product information can be made available in a standardised, reusable form that supports AI-assisted specification and asset information workflows. The library demonstrates the principle that high-quality, consistently structured information, when made accessible through open or standardised interfaces, enables downstream digital processes including GenAI-assisted design, specification, and handover workflows. Construction organisations building Hub knowledge bases that include product specification information may wish to explore how NBS library content can be incorporated into their information estate in a structured and reusable way.
The resources in this group address the governance, risk management, and data protection frameworks that define safe and accountable AI use for organisations operating in the UK and internationally. These are not optional readings for governance teams alone. Every construction professional who uses, configures, or procures GenAI tools needs at minimum a working familiarity with the professional standards and regulatory guidance in this group. The RICS professional standard is obligatory for RICS members and regulated firms from March 2026. The ICO guidance reflects statutory obligations under UK GDPR. The ISO and NIST frameworks provide the governance vocabulary that clients, insurers, and auditors increasingly use when assessing AI risk.
This is the RICS landing page for the Responsible Use of Artificial Intelligence in Surveying Practice professional standard, published in November 2025. The standard came into effect on 9 March 2026 and is mandatory for all RICS members and regulated firms where AI is used in a way that has a material impact on the delivery of surveying services. The landing page provides access to the standard itself, a response to consultation document explaining the basis for the standard's conclusions, a set of sector-specific case studies, a pre-recorded webinar, and a set of frequently asked questions that address common points of uncertainty about the standard's application.
The standard was developed by an expert working group co-chaired by Sophia Adams-Bhatti, a public policy and regulation expert, and Christopher de Gruben FRICS, Director at Artefact and a leading figure in AI adoption in property. The group included practising chartered surveyors from across the built environment, a commercial barrister specialising in technology law, and risk management specialists, providing a genuinely cross-disciplinary foundation for the standard's requirements.
The standard addresses six substantive areas: knowledge requirements for using AI in surveying practice; practice management including data governance, system governance, and risk management; AI procurement and due diligence processes; output reliability and assurance protocols; client communication and transparency requirements; and AI system development guidance for firms creating their own AI solutions. These areas map closely to the Hub's own governance framework and competency pathway content, making the RICS standard an important reference point for construction organisations assessing whether their Hub deployment practices meet the requirements of professional regulation.
Several points from the standard and its accompanying FAQ are particularly relevant to Hub users. The standard applies only to AI use that has a material impact on surveying service delivery, which means that administrative uses such as managing room bookings or drafting internal emails are likely outside scope, while uses that affect professional outputs such as valuations, cost assessments, risk reports, or contract administration communications are clearly within scope. The standard requires that written decisions about output reliability be prepared by or under the supervision of a named, appropriately qualified surveyor, a requirement that directly supports the Hub's emphasis on human oversight and professional accountability. The standard also requires that AI use be addressed in terms of engagement with clients, reinforcing the Hub's guidance on transparency and disclosure.
The landing page also includes an important caution about cybersecurity: the standard notes that AI can increase cybersecurity risk through its capacity for pattern recognition and mimicry, which are tools that can be used in hacking, and through the potential for AI deployment to create system vulnerabilities exploitable by third parties. The real estate sector is specifically identified as an emerging target for cybersecurity attacks, given the financial information and identity documents held by real estate and surveying firms.
This is the full text of the RICS professional standard, published in September 2025 and available as a freely downloadable PDF from the RICS website. The full standard contains the normative requirements that RICS members and regulated firms must comply with, the definitions and scope provisions that determine how those requirements apply, the supporting guidance that helps practitioners interpret and implement the requirements, and the case studies that illustrate how the standard applies across different surveying practice areas including land and natural resources, valuation, residential property, commercial property, construction, building surveying, and general practice.
The construction case study is of particular relevance to Hub users. It addresses the use of AI in quantity surveying, project management, and contract administration contexts, and illustrates how the standard's requirements for output reliability, professional oversight, and client communication apply in these specific settings. The case study acknowledges the genuine value that AI tools can add in cost management and document analysis workflows while maintaining clear expectations about the professional accountability that must accompany AI-assisted outputs.
The standard's treatment of AI hallucinations, failure modes, and limitations is clear and practically oriented. It defines hallucinations as cases where AI generates false or invented output and notes that these can be viewed as either a limitation or a failure mode depending on their nature and extent. It identifies common failure modes including improper access controls, data corruption, bias amplification, misinterpretation of instructions, and human-in-the-loop bypass, providing practitioners with the vocabulary to describe and manage AI risks in professional contexts. The standard is 34 pages in total and is essential reading for any RICS member or regulated firm deploying the Hub in a professional surveying context.
https://nvlpubs.nist.gov/nistpubs/ai/nist.ai.100-1.pdf
The NIST AI Risk Management Framework, published in January 2023 by the National Institute of Standards and Technology within the US Department of Commerce, is one of the most widely referenced AI governance frameworks in the world. Although developed in the US context, it has been adopted as a reference framework by organisations globally, including many UK construction companies and their clients. Its four core functions, Govern, Map, Measure, and Manage, provide a structured approach to AI risk that is grounded in established risk management principles and designed to be adaptable across different organisational sizes, sectors, and AI deployment contexts.
The Govern function covers the policies, processes, procedures, and practices that define how an organisation approaches AI risk across all of its AI deployments. The Map function covers the identification and classification of AI risks in context, taking into account the specific use case, deployment environment, and potential impacts on individuals and organisations. The Measure function covers the tools and methods used to assess and monitor AI risks, including evaluation frameworks, testing methodologies, and performance metrics. The Manage function covers the responses to identified and measured risks, including risk treatment decisions, mitigation actions, and residual risk acceptance.
For UK construction organisations, the NIST AI RMF provides a useful complement to the more sector-specific and professionally grounded RICS standard. Where the RICS standard focuses on the obligations of professional practice within surveying, the NIST framework provides a broader organisational governance vocabulary that is applicable across all of an organisation's AI deployments, not only those that are specifically surveying related. Larger construction organisations that have enterprise-level AI governance requirements, particularly those operating internationally or under client contractual requirements that reference the NIST framework, will find it essential. The framework is freely available as a PDF download from the NIST website.
https://www.iso.org/standard/42001
ISO/IEC 42001:2023, published in December 2023, is the world's first international standard for artificial intelligence management systems. It specifies requirements for establishing, implementing, maintaining, and continually improving an AI management system within organisations that develop, provide, or use AI-based products or services. It is designed to be applicable across all industries and all organisational sizes, using the familiar Plan-Do-Check-Act management system methodology that underpins other well-established ISO management standards including ISO 9001 for quality management and ISO 27001 for information security management.
The standard was developed by ISO/IEC Joint Technical Committee 1, Subcommittee 42, which is specifically dedicated to artificial intelligence. It addresses the unique challenges that AI poses to organisational governance, including ethical considerations, transparency requirements, the need for ongoing monitoring of AI system behaviour, and the management of AI-specific risks such as bias, hallucination, and unintended automation of consequential decisions. The standard provides a framework for managing AI risks and opportunities in a structured, documented, and independently auditable way.
ISO/IEC 42001 is relevant to Hub deployments in several specific ways. Its requirements for AI risk assessment, impact assessment, and lifecycle management provide a governance vocabulary and process framework that construction organisations can use to structure their approach to Hub deployment. Its requirements for documentation, record-keeping, and continual improvement align with the audit trail and evaluation functions described in Section 12 of this document. And its status as an international standard means that conformance with it provides a credible, externally recognisable signal of governance maturity that can be relevant in procurement and client relationship contexts.
The standard is available from the ISO store at CHF 225 for the PDF version. ISO also offers a bundled AI and information security management package combining ISO/IEC 42001 with ISO/IEC 27001 at a discounted price, reflecting the close relationship between AI governance and information security management. Construction organisations that are implementing both information security and AI governance frameworks are recommended to consider both standards together, given the significant overlap in their organisational infrastructure requirements.
The Information Commissioner's Office Guidance on AI and Data Protection is the primary regulatory guidance document for UK organisations on the application of UK GDPR to AI systems. It is maintained by the ICO as a living document and was most recently updated in March 2023, with further updates signalled in response to ongoing industry engagement and regulatory developments. The Data (Use and Access) Act 2025 received Royal Assent on 19 June 2026, and its provisions affecting data protection law are now in force; organisations should monitor the ICO website for updated guidance reflecting the Act's implications.
The guidance is structured around the data protection principles established in UK GDPR and applies them systematically to AI systems across their full lifecycle. It covers accountability and governance implications of AI, including Data Protection Impact Assessment requirements for high-risk AI processing; transparency obligations, including requirements for explaining AI-assisted decisions; lawfulness requirements, including the identification of appropriate lawful bases for personal data processing in AI workflows; accuracy considerations in AI systems; and fairness in AI, which was significantly expanded in the March 2023 update.
The fairness section is particularly substantive and addresses the distinction between data protection fairness and algorithmic fairness, the sources of bias that can lead to unfair AI outputs, and the technical and organisational measures available to mitigate algorithmic bias across the AI lifecycle. An annex on fairness in the AI lifecycle addresses each stage from problem formulation through to decommissioning, identifying the ways in which design and data decisions at each stage can introduce or compound unfairness. This content is directly relevant to construction AI applications where training data drawn from historical project records may embed historical biases in procurement, cost estimation, or risk assessment.
For Hub operators and users, the ICO guidance defines the baseline data protection compliance requirements for all AI processing that involves personal data. Construction project information frequently contains personal data, including names of project team members in correspondence, personal contact details in health and safety records, and individual performance information in workforce management systems. Understanding which AI processing activities require a lawful basis, which require a Data Protection Impact Assessment, and what transparency obligations apply to AI-assisted outputs is essential for operating within the law and maintaining the trust of project stakeholders.
The resources in this group address the policy and procurement framework governing public works projects in the UK. Construction organisations working on publicly funded projects, whether as main contractors, consultants, or specialist subcontractors, operate within this framework and need to understand how it affects the expectations placed on them and the decisions they make about technology adoption, information management, and supply chain engagement. As AI tools become more embedded in construction practice, the Construction Playbook and related guidance will increasingly provide the context within which AI procurement and deployment decisions in the public sector are assessed.
https://www.gov.uk/government/publications/the-construction-playbook
The Construction Playbook was first published by the Cabinet Office in December 2020 and most recently updated in July 2023. It sets out the key policies and guidance governing how central government departments and their arms-length bodies assess, procure, and deliver public works projects and programmes. All central government departments are expected to follow the Playbook's fourteen key policies on a comply or explain basis, making it the central policy document for public sector construction procurement in England. Equivalent frameworks apply in Scotland, Wales, and Northern Ireland under their respective devolved administrations.
The Playbook captures commercial best practice and sets out the government's expectations for how contracting authorities and suppliers, including the supply chain, should engage with each other. Its fourteen key policies address areas including outcome-based procurement, should-cost modelling, risk allocation, contract standardisation, and the use of modern methods of construction. The accompanying guidance notes published alongside the updated Playbook in September 2022 cover modern methods of construction, longer-term contracting, net-zero carbon and sustainability, and market and supply chain engagement.
The Playbook's relevance to GenAI in construction lies primarily in the procurement and should-cost modelling domains. The Playbook's emphasis on evidence-based cost assessment and transparent risk allocation creates a context in which AI-assisted cost intelligence and risk analysis tools are potentially valuable, but also in which the basis and limitations of AI-generated cost figures and risk assessments must be clearly understood and disclosed. Construction organisations using GenAI tools to support bids and cost plans for public sector projects need to ensure that their use of AI is consistent with the transparency and auditability expectations of public procurement, and that AI-generated outputs can be defended under should-cost scrutiny.
The Playbook is freely available from the GOV.UK publication page, which links to the main document and all associated guidance notes. The main Playbook document is 89 pages in PDF format. Organisations working regularly on public sector construction should maintain current familiarity with the Playbook and monitor for further updates, as the government has signalled its intention to continue developing the framework in response to emerging issues including digital and AI adoption.
This is the direct link to the current version of the Construction Playbook PDF, updated in September 2022. The document is published by the Cabinet Office under the Open Government Licence version 3.0, making it freely reproducible with appropriate attribution. The 89-page document sets out the fourteen key policies that govern public works procurement, together with the commercial best practices and expectations that support each policy. It is the working reference for procurement teams, project managers, and commercial advisers engaged in public sector construction, and should be read in conjunction with the accompanying guidance notes that address specific policy areas in greater depth.
For GenAI practitioners, the Playbook is most directly relevant in its treatment of information requirements, data sharing, and supply chain engagement. The model clause for subsurface data sharing that accompanies the Playbook is an example of how the government has used contractual mechanisms to promote data sharing in construction, a principle that is directly applicable to the question of how project information is made available for AI processing in publicly funded projects. The Playbook's emphasis on longer-term contracting relationships and framework agreements also creates opportunities for the sustained development of shared AI knowledge bases across programmes of work rather than project by project, which is a more efficient and more governable approach to GenAI deployment in the public sector context.
The resources in this group provide access to the technical documentation for the major cloud platforms through which language model capabilities are most commonly accessed and deployed in enterprise construction contexts. These are the platforms through which organisations build RAG systems, deploy conversational AI interfaces, and access foundation model APIs for custom integration work. The documentation pages are the authoritative technical references for each platform and should be the first port of call for data engineers and GenAI architects working on Hub technical integration.
A note on platform evolution: the GenAI platform landscape is changing rapidly, and the documentation for each platform is updated frequently as new capabilities are released. Readers should access the live documentation pages rather than relying on any static description of platform capabilities, including the descriptions provided here, which reflect the state of each platform as of the time of writing in June 2026.
https://docs.aws.amazon.com/bedrock/
Amazon Bedrock is a fully managed AWS service that provides access to foundation models from multiple providers through a single API, without requiring organisations to manage the infrastructure needed to run large language models themselves. The Bedrock documentation covers the full range of capabilities available through the service, from basic inference and model access through to the more complex builder tools that support enterprise AI application development.
The key capabilities documented include knowledge bases, which implement retrieval-augmented generation by connecting Bedrock models to organisational data sources and enabling them to generate grounded responses based on retrieved document content; agents, which enable the construction of autonomous AI systems that can plan and execute multi-step tasks using defined tools and data sources; flows, which allow the orchestration of multiple Bedrock features and AWS services into coherent generative AI workflows; and guardrails, which implement configurable safeguards based on responsible AI policies and use-case specific content restrictions.
The knowledge bases capability is the most directly relevant to Hub deployments. It supports the ingestion of documents from multiple source types, handles chunking and embedding automatically using integrated vectorisation, and enables retrieval using keyword, semantic, vector, and hybrid search strategies. The guardrails capability is relevant to the governance requirements described throughout this document, enabling organisations to implement content policies that prevent AI outputs from including harmful, inaccurate, or out-of-scope content. The agents capability is relevant to the agentic workflow scenarios described in Sections 10 and 12, enabling the construction of autonomous AI systems that can take structured actions on behalf of users within defined permission boundaries.
Bedrock supports models from Amazon, Anthropic, Meta, Mistral, and other providers, allowing organisations to select models appropriate to their specific use cases without changing their application architecture. Cross-region inference allows requests to be routed across multiple AWS regions, improving availability and supporting data residency requirements. Security configuration guidance covers identity-based access policies, managed policies, and service roles, providing the technical foundation needed to implement the access control requirements described in the Hub's governance framework.
This Microsoft Learn documentation page describes the Azure OpenAI On Your Data capability, which enables organisations to run GPT models against their own enterprise data using a retrieval-augmented generation approach managed through the Azure platform. It is important to note, however, that Azure OpenAI On Your Data has been deprecated and is scheduled for retirement on 14 October 2026. Microsoft has stopped onboarding new models to this service and recommends that organisations migrate existing workloads to the Foundry Agent Service with Foundry IQ.
Despite the deprecation notice, the documentation remains a valuable technical reference for understanding how RAG systems are implemented at enterprise scale, because the concepts it describes, including document ingestion, chunking, embedding, retrieval, and response generation, are transferable to the current and future Azure AI platform offerings. The documentation provides detailed coverage of the RAG pipeline, explaining how user queries are processed through intent generation, document retrieval, filtration and reranking, and response generation stages. This four-stage pipeline description is a clear and accurate representation of how well-implemented RAG systems work regardless of the specific platform.
The documentation's treatment of chunk size configuration is particularly useful for practitioners building construction document knowledge bases. It explains that the default chunk size of 1,024 tokens may not be optimal for all document types, and provides guidance on when smaller chunks of 256 or 512 tokens may improve retrieval granularity and when larger chunks of 1,536 tokens may better preserve contextual information. This guidance translates directly to the RAG cookbook and starter kit materials described in Section 12 of this document.
The documentation also covers document-level access control, which allows search results to be filtered based on user identity and group membership, ensuring that AI-generated responses draw only on documents that the requesting user is authorised to access. This capability is directly relevant to the access control architecture of the Hub's private portal, where different users within an organisation may have different document access rights based on their project role or security clearance. Organisations planning Hub deployments that involve sensitive project information should review the access control architecture carefully against their specific data protection and security requirements.
For current Azure AI deployments, practitioners should navigate to the Microsoft Foundry portal and the Foundry Agent Service documentation, which represents the current strategic direction for Microsoft's enterprise GenAI platform rather than the deprecated On Your Data service.
https://docs.cloud.google.com/vertex-ai/docs
Google Vertex AI is Google Cloud's unified platform for machine learning and generative AI, providing access to Google's Gemini family of models alongside a range of tools for building, deploying, and managing AI applications at enterprise scale. The Vertex AI documentation covers the full range of platform capabilities relevant to enterprise AI deployment, from model access and fine-tuning through to the search and retrieval tools, agent frameworks, and evaluation infrastructure that support production GenAI applications.
The Vertex AI Agent Builder is the component of the platform most directly relevant to Hub deployment scenarios. It provides managed search and retrieval capabilities that can be configured against enterprise document repositories, enabling organisations to build RAG-based question-answering and document synthesis applications without managing the underlying vector search infrastructure. The platform supports ingestion from Google Drive, Google Cloud Storage, BigQuery, and third-party data sources, and provides both a visual configuration interface and a programmatic API for more sophisticated integration scenarios.
Vertex AI also provides access to Google's multimodal models, which can process images and documents alongside text. For construction applications, this multimodal capability opens use cases in drawing review, site photograph analysis, and document image extraction that are not supported by text-only model deployments. The platform's evaluation tools include automated metrics for assessing response quality, groundedness, and citation accuracy, which align with the evaluation framework requirements described in Section 10 of this document. UK construction organisations with existing Google Cloud infrastructure investments will find Vertex AI a natural home for Hub technical integration work.
https://platform.openai.com/docs/api-reference/introduction
The OpenAI platform provides the API through which OpenAI's language models, including the GPT-4o family and the o-series reasoning models, are accessed programmatically. The OpenAI API is one of the most widely used interfaces for GenAI application development globally, and many construction technology integrations that use large language models do so through this API, either directly or through orchestration frameworks such as LangChain that abstract the API interaction. Note: the OpenAI documentation page blocked automated access at the time of writing, so the description here is based on known published documentation and public technical resources rather than a live page read.
The core API provides access to the chat completions endpoint, which is the primary interface for conversational and document synthesis applications. It supports structured output formats, function calling for tool use and agentic workflows, and vision inputs for multimodal applications. The API also provides access to the embeddings endpoint, which generates vector representations of text for use in semantic search and retrieval systems, and the fine-tuning endpoint, which allows organisations to customise models on their own data for domain-specific applications.
For Hub technical integration, the OpenAI API is most relevant as the model access layer in custom RAG implementations where organisations are building their own retrieval and orchestration logic rather than using a managed platform service. Organisations choosing this approach gain greater control over the retrieval and generation pipeline but take on more engineering responsibility for performance, reliability, and governance. The API supports both standard and enterprise access tiers, with the enterprise tier providing additional data privacy assurances, including a contractual commitment that data submitted through the API will not be used for model training.
The Responses API, introduced in 2025, and the Assistants API provide higher-level abstractions for building stateful conversational agents and multi-turn applications, reducing the engineering overhead of managing conversation history and tool state in complex agentic workflows. These abstractions are relevant to the agentic construction workflow scenarios described in Section 10 and Section 12 of this document. Practitioners planning OpenAI API integrations should consult the current platform documentation directly, as the API surface has evolved significantly over the past twelve months and will continue to do so.
The resources compiled in this section represent a curated starting point rather than an exhaustive bibliography. The construction and AI landscapes are both moving rapidly, and new standards, guidance documents, and platform capabilities are emerging continuously. Users of this list are encouraged to treat it as a living reference that requires regular review rather than a fixed catalogue.
For each resource, the appropriate level of engagement will depend on the user's professional role and stage of GenAI adoption. The ISO 19650 standard, the RICS AI professional standard, and the ICO guidance on AI and data protection are essential reading for any professional operating in a governance or senior practitioner role. The platform documentation resources are primarily relevant to data engineers and GenAI architects working on technical implementation. The UK BIM Framework and buildingSMART standards resources are relevant across a wide range of roles, from information managers and BIM coordinators to project directors responsible for digital delivery strategy.
Where resources are referenced in other sections of this document, the cross-references are designed to help users move between the conceptual guidance in the Hub and the primary sources that underpin it. The principle throughout is that Hub guidance should help practitioners engage confidently with primary sources rather than substituting for them. A quantity surveyor who reads Section 10's description of the RICS AI professional standard and then accesses the full standard through the link provided is better placed to apply it in their practice than one who relies on a summary alone. The same principle applies to all resources in this section: the Hub is a guide to the landscape, not a replacement for the authoritative sources that define it. Further updates to this resource list, including new standards, regulatory guidance, and platform documentation, will be published on the Hub's resource library page at iso.org, rics.org, and ico.org.uk as they become available.