RAG search on construction documents works by first retrieving the passages most relevant to a question from across your drawings, reports, specifications and correspondence, then giving those passages to a language model that writes an answer and cites the files it used. Retrieval augmented generation (RAG) keeps answers grounded in your own documents. AI Insights in the Utopiaverse™ applies RAG, together with optical character recognition (OCR), to the connected record in your own Azure tenancy.
The two halves of RAG
RAG has two parts. The retrieval part searches an index of your documents for the passages most likely to answer the question. It uses the meaning of the text as well as keywords, so a question about "water ingress" can find passages about "leaks" or "moisture". Metadata filters narrow the search, for example to a project, discipline or status.
The generation part passes those passages to a language model, which writes an answer based on them. Because the model is working from retrieved passages, it can cite exactly which files and pages support each point. That is the key difference from asking a general AI tool, which answers from what it learned during training and cannot show where an answer came from.
Why construction documents are a special case
Construction documents are harder for RAG than typical office files. Many are scanned PDFs or drawings where the useful text is inside an image, so OCR is essential. Documents exist in many revisions, and an answer drawn from a superseded revision can mislead. Metadata such as status, revision and discipline carries meaning that matters to engineers. The same project may hold content in several common data environments (CDEs), each with its own structure.
A RAG system for construction needs to handle all of these. The Utopiaverse™ does so by bringing content from every CDE into your Azure with version history and metadata preserved, mapping that metadata to a common schema aligned with ISO 19650, and running OCR across scanned content before indexing.
How to set up RAG search on your documents, step by step
Decide the questions you want answered and the projects and platforms that hold the answers.
Bring content into your Azure with CDE Backup™, CDE Sync™ or CDE Archive™.
Agree the common metadata schema so filters work across platforms.
Let the content be processed with OCR and indexed.
Test AI Insights with real questions and check that citations point to the right files and revisions.
Refine metadata or scope where answers are weak, then widen access.
Getting reliable answers
RAG answers are only as good as what is retrieved. A few practices make a big difference. Include the full record, because missing platforms mean missing answers. Keep metadata consistent, so filters by project, status and discipline behave predictably. Preserve version history, so the system can favour current revisions and explain changes. Encourage people to open the cited sources, especially for decisions with contractual weight. With those in place, RAG becomes a dependable way to find and understand information across a large project record.
RAG compared with keyword search
Keyword search returns a list of documents that contain the words you typed, and leaves you to open each one and work out the answer. RAG goes further by reading the relevant passages and summarising what they say, while still pointing to the documents. That saves time when the answer is spread across several files, such as a design change discussed in correspondence, recorded in meeting minutes and reflected in a revised drawing. Keyword and metadata search still have a place, especially for people who know exactly which document they need, and AI Insights supports both.
Where RAG runs and why it matters
Many RAG tools send documents to an external service to be indexed. For construction and infrastructure organisations, that raises questions about confidentiality, data residency and critical infrastructure obligations. AI Insights runs natively on Azure inside your own tenancy, in the region you choose, so documents are not copied to a third party to be indexed. Utopia Digital holds SOC 2 Type II and Cyber Essentials. See the Security Trust Portal and the global data region map.
RAG search and the Utopiaverse™
RAG is how AI Insights turns the Utopiaverse™ into answers. CDE Sync™, CDE Backup™, CDE Migrate™ and CDE Archive™ build the connected record in your own Azure, and AI Insights retrieves, reasons and cites across it. See also How Can I Enable AI to Read My Common Data Environment? and How Do I Search Across Multiple CDEs at Once?
See RAG search on real project documents in the monthly live demo, or talk to the team about your record.
