Project Knowledge Assistant
Ask a question and get the answer straight from the firm's own specs, head contract, ITPs, and standards — with the source passage, a link, and a confidence flag — instead of interrupting whoever last touched the file.
A private retrieval assistant reads only your approved project documents, answers with the quoted passage and a link, flags its confidence, and refuses when the answer isn't in the corpus.
Where this works well
The slow, invisible cost this makes visible is the interruption tax: every time someone needs the minimum concrete cover, the right ITP revision, or what a GC21 clause actually obliges, they break someone else's concentration to ask, and the answer arrives in minutes or hours — or wrong. A project of any size accumulates a spec running to dozens of revisions, an executed head contract, inspection and test plans, and referenced standards like AS 3600:2018, and the knowledge of where the answer lives sits in two or three experienced heads.
It earns its keep when the corpus is genuinely the source of truth — a current project spec, the executed contract, live ITPs, the standards you actually build to — and the questions are factual and document-answerable. The people who benefit most are site engineers, document controllers, and project managers on mid-to-large jobs where the file count is high enough that no one holds it all, and where the same questions recur across a long build. It turns "ask whoever remembers" into a cited answer in seconds, with the source clause attached so the asker can verify it themselves.
Where it works badly
It is confidently wrong when the corpus is stale or ambiguous. Index last month's spec instead of the current revision and it will quote superseded reinforcement detail at 94% confidence — the confidence score measures how well the answer is grounded in the indexed text, not whether that text is still correct. Leave three revisions of the same document in one folder with no superseded marking and it can retrieve the wrong one. Scanned drawings that were never OCR'd are invisible to it, so it will answer "that isn't in your documents" when in fact it is — just not as searchable text.
It is also the wrong tool for judgement questions. "Should we vary the pour given the forecast?" is not a retrieval question; the documents inform that call but don't make it. And on a small, single-project job where everything fits in one folder and one person knows it, the indexing effort may not pay back.
The honest test: pick a question your team asks weekly and check whether one document, in one current revision, definitively answers it. If yes, this pattern helps. If the real answer is "it depends, ask the engineer," no retrieval tool fixes that.
What it doesn't do — and shouldn't
It surfaces and quotes; a person decides what is authoritative and what to build to. It will tell you the spec says 30 mm cover and the standard agrees — and in the same breath flag that you should confirm the exposure classification for any slab-on-ground sections with the project engineer before the pour. That amber caution beat is deliberate. The assistant does not approve a pour, sign an ITP, interpret a contractual entitlement, or resolve a conflict between two documents. Where the spec and a standard disagree, it shows you both and stops — the reconciliation is an engineering and contractual decision, not a retrieval one.
The human-in-the-loop boundary matters here because the consequences are physical and contractual: getting cover, lap lengths, or a GC21 notice period wrong has safety, durability, and liability consequences that a confidence bar cannot carry. The accept / refine / flag-as-wrong step, and the escalation to a named person for out-of-scope questions, are the point — not friction to be removed.
What your data has to look like for this to work
It needs a governed corpus: one authoritative copy of each document, at its current revision, with superseded versions removed or clearly labelled. Each document carries enough structure that a clause, an ITP step, or a spec section can be cited precisely — section numbering that survives ingestion, not a flat wall of text. Scanned PDFs and marked-up drawings need OCR so their text is retrievable. Access roles need to be defined so the corpus respects who is allowed to see the head contract or tender pricing. And it needs to be re-indexed when documents change, so the newest revision is what gets answered from.
Most firms have some of this and not all of it. The spec might be clean while the drawings register is three weeks behind; the contract is governed while past tenders sit unsorted across old project folders. Getting to that governed, current, searchable state is usually the bigger and more valuable piece of work — and it is a question of how documents are captured and controlled, not of buying a new tool. That is the work we help with, and it is what makes the retrieval layer trustworthy rather than merely fast.
Could it give me a wrong concrete cover or clause and I'd build to it?
Every answer carries the quoted source passage, a link to the document, and a grounding-confidence flag, and the demo shows an amber caution beat telling you to confirm exposure classification with the project engineer before a pour. The assistant surfaces what the spec and the standard say; a competent person still confirms it against the as-built condition and signs off. It is built to make checking faster, not to remove the check.
Our project files are a mess — duplicate revisions, scanned PDFs, half the drawings register out of date. Will it still work?
Only as well as the corpus you point it at. If three revisions of the spec sit in the same folder with no superseded marking, it can quote the wrong one with high confidence. Getting to one governed source per document — current revision, superseded versions removed or labelled, scans run through OCR so the text is searchable — is usually the real first job, and it is the work we help with before the AI layer earns its keep.
Does this replace our document controller or the project engineer?
No. The document controller still owns which revision is current and what is authoritative; the project engineer still makes the engineering judgement. The assistant recaptures the time those people lose answering the same question by hand and fielding interruptions, and redirects it to the work that needs their judgement. It draws the line deliberately: it surfaces and quotes, a person decides.
How current does the indexed content have to be?
Current enough that the newest authoritative revision is the one indexed and superseded ones are out. The corpus shows each document's chunk count and last-updated date so you can see what it is answering from. If you re-index on a stale snapshot — last month's spec, a head contract amended since execution — it will answer confidently from the old text. Re-indexing on document change is part of the setup, not an afterthought.
Where does our data go? These are commercial contracts and tender prices.
It stays inside your own tenancy — your SharePoint or Drive — and the retrieval runs against your private corpus, not a public model trained on your files. Access roles mean a site labourer and a commercial manager don't see the same documents. Nothing in the head contract or a past tender leaves your environment to answer a question.
What happens when the answer genuinely isn't in our documents?
It says so and refuses to guess, rather than inventing a plausible number. The demo's second question — annual leave for site labourers — is answered by declining, because that lives in the HR system, not the project knowledge base, and then offering to escalate to a named human. A logged gap also lets the document owner decide whether that material should be indexed.
Estimated build: 3–4 weeks. Most of it is template work we've already done.
Fixed scope, fixed price, fixed dates.
The cost band reflects the engagement shape, not a per-feature line item. We work on fixed scope, fixed price, fixed dates — see the services catalogue for what falls inside each band.
Considering this for your org?
The honest place to start is a bite-sized first piece — one contained change, low risk. Tell us where it hurts; we’ll play it back, scope it, and show you what’s possible.