Emerge Insights
Tender Drawing Sets to a Quantity Surveyor's Workbook
A multi-tenant platform that classifies a builder's tender drawing set by trade and extracts quantities into a tender-ready Excel workbook, labelling every number by how it was obtained.
- Industry
- Construction / Quantity Surveying
Quantity take-off from tender drawings is slow, expensive specialist work. Automating it naively produces numbers a quantity surveyor cannot verify, which makes them unusable in a document they are professionally accountable for.
- Take-off cost runs into four figures per estimate
- A vision model alone produces values nobody can check
- Drawing sets mix trades, and each sheet type needs different logic
- Printed values, computable geometry and genuine estimates were treated identically
- Holding several builders' tender sets requires isolation from day one
A multi-tenant pipeline that classifies each sheet by trade, extracts quantities through a strict Read then Computed then Inferred preference order, and writes a tender-ready Excel workbook that labels the provenance of every number.
- Classify every sheet in the set by trade before extraction
- Prefer values read verbatim from the PDF text layer over anything derived
- Compute from vector geometry in our own code where a value is not printed, keeping it deterministic
- Fall back to vision inference only when a value is neither printed nor computable, and flag its confidence
- Mark the provenance tier of every value in the Excel output
- Build multi-tenant authentication and isolation into the first version
How it was built
Taking off quantities from a tender drawing set is expensive specialist work, and automating it with a vision model alone would be worse than useless. The output goes into a tender someone is accountable for, so a value nobody can verify has no value.
Values are produced in a fixed order of preference. Read means lifted verbatim from the PDF's own text layer, because it was printed on the drawing. Computed means derived from the file's vector geometry by our own code, with no model in the path, so it is repeatable to the millimetre. Inferred means measured or counted by a vision model, used only when the value is not printed and cannot be computed.
The Excel marks which tier each number came from, and every inferred value carries a confidence flag. The surveyor knows what to trust on sight, which is the whole product. This preference order later became a studio-wide standard applied to retrieval answers and price lookups as well.
Every sheet in the set is classified by trade first, covering layout, electrical, plumbing, finishes and elevations, so each one is read by logic that expects what is on it. Room areas, fixture counts, wall lengths and opening dimensions all mean different things depending on the sheet they came from.
Authentication and per-tenant isolation were built in at the start rather than retrofitted, because a platform that holds one builder's tender drawings alongside another's cannot add isolation later and be believed.
- Four-figure quantity-surveyor cost per estimate
- Manual take-off across every sheet in the set
- No distinction between a printed value and an estimate
- Nothing to verify an automated number against
- Cost per estimate reduced to a few euros
- Sheets classified and read by trade automatically
- Every number labelled Read, Computed or Inferred
- Inferred values carry an explicit confidence flag