RAG turns past project evidence into a source-backed starting point

Past project knowledge is only useful when a team can find the right evidence quickly.

Construction teams have documents, decisions, permits, policies and correspondence spread across projects. A search result without context can be just as dangerous as no result. The team needs to know why a document is relevant, which project it came from and whether it still applies.

We added retrieval-augmented generation, or RAG, to the construction knowledge system. RAG retrieves relevant source documents before producing an answer. The source project and document remain visible, so the next team can use the earlier work as a starting point and still verify the current situation.

Client background

Our client is a large construction company delivering public bridge contracts, housing developments and other multi-site projects. Requirements differ by municipality, project type, contract, site and date. The business needed company-wide knowledge transfer without flattening those differences.

The relevant information already existed. The problem was that people had to know which folder, inbox or colleague to search before they could use it.

Business challenge

Traditional search returns words. Project teams need an answer connected to operational context. When a project manager asks how a similar development handled a requirement, the useful result is not every document containing the same phrase. It is the relevant decision, its source, its date, the project that produced it and the person who can explain it.

Without that context, teams face two bad choices:

• repeat the research because the previous work cannot be found with confidence;
• or reuse an old document as if it were a current approval.

Neither creates good governance. The first pays twice for knowledge. The second creates a compliance risk by hiding the difference between experience and a current requirement.

Implementation

We structured project records so that documents carry the context needed for retrieval. The system connects a source document to the project, municipality, contract, work package, date, decision and responsible person where that information is available.

RAG then works on top of that structure. When someone asks a question, the system retrieves relevant source material before generating the response. The answer can point the team back to the document and project that support it instead of presenting an unsupported summary.

The clearest example is City A, Project X to City A, Project Y. The new project can ask what the earlier project did and receive the relevant requirements, decisions and evidence. The earlier source stays visible. The team can then compare the current site and contract, ask the responsible person for context and verify that the local requirement has not changed.

RAG is therefore one part of a wider governance model. It helps the team retrieve. The project structure tells the team what the retrieved material means. Human responsibility remains in the loop for current compliance and project approval.

Why it was difficult

Construction knowledge is highly contextual. A document can be relevant because it concerns the same municipality, but not the same project type. A bridge decision is not automatically the rule for a housing development. A previous permit may be informative without being current.

The system had to make the relevant evidence easy to find without creating false certainty. Source visibility, project context and dates are what turn retrieval into a useful operating control.

Value delivered

• RAG retrieves relevant project documents before an answer is produced.
• The source project and document remain visible to the person using the answer.
• City A, Project X can provide a structured starting point for City A, Project Y.
• Teams can search by municipality, project type, contract and work package.
• Previous decisions and evidence become reusable without becoming automatic approval.
• Compliance and delivery professionals can verify the current requirement from the original source.
• Knowledge transfer becomes part of company governance rather than an informal search for memory.

The outcome is not “the system knows construction”. It is that the company can find the relevant work it has already completed and see enough context to use it responsibly.

What this means for your business

If your organisation has tried search or AI and still cannot trust the result, the missing layer may be source and context. Retrieval should show where an answer came from and why that source belongs to the question.

That applies to compliance libraries, case management, policy work and operational playbooks. The useful question is not whether the system can generate an answer. It is whether the next person can find the evidence and judge if it still applies.

Related: Knowledge systems · Applied research · Construction companies