How to Use Teraquant: A Field Guide for QS Teams

Teraquant is built around one idea: AI should make a quantity surveyor faster, never replace their judgement. Every number the AI proposes stays a proposal — linked to the exact drawing region it came from — until a human reviewer signs off. This guide walks through the full workflow, from your first upload to a signed-off, exportable quote.

Getting started

Sign in at teraquant.ai and create a project, or open an existing one from your project list. Upload your MEP drawing PDFs (plumbing & drainage, fire services, HVAC, electrical) and the project's SOR/BQ spreadsheet. Teraquant reads each drawing's filename to infer its discipline and routes it straight to the matching SOR sheet — no manual tagging needed, and it never silently falls back to the wrong sheet.

Extracting quantities from drawings

Open any drawing and you land on the Drawing Summary — an AI-generated read of what's on the page, plus a quantity match rate against your SOR. From there you have several ways to extract quantities:

• Region-select OCR — draw a box around anything (a label, a dimension, a symbol) and let the AI read it.
• Guided legend takeoff — for repeated symbols like sanitary fittings or valves, box the drawing's legend once and the AI counts every occurrence of each symbol type across the whole page.
• AI geometric length takeoff (幾何量度管長) — a Schedule C / inspector action that runs server-side geometric takeoff for metre (`m`) SOR rows: the AI traces pipe/duct centerlines and Teraquant converts them to real-world length using the drawing scale (1:N). This is not a guess; it is a geometric calculation on the server.
• Manual measure mode — a separate viewer-toolbar tool: you click polylines yourself on the PDF, and the app applies the page scale (or your calibration) to report length. Use it for spot checks or runs the AI missed. It is not the same feature as AI geometric length takeoff.

If a page is marked "not to scale," Teraquant says so honestly instead of fabricating a length — a number you can't trust is worse than no number at all. Text OCR alone does not produce pipe lengths; use AI geometric takeoff or manual measure when you need metres.

Reviewing before anything counts

Every extracted value shows up in the Claim Review Panel with its status (draft, needs clarification, accepted, rejected) and confidence score. While a claim is still a draft — you do not have to accept first — you can run Match to SOR: the panel suggests the top candidate rows and flags unit mismatches before you confirm. You can also accept, reject, or request clarification at any time. Every correction you make is preserved alongside the AI's original proposal, so nothing is ever silently overwritten.

From accepted claim to priced quote

Once you accept a claim and confirm its SOR match, Teraquant looks up the latest rate from your cost database (`cost_items`) and computes the line total automatically. If unit rates are missing — empty cost rows, or formula-shell SOR rates with no filled prices — exports stay unpriced until you add rates on the project costs page. When you're ready, export a priced XLSX with formulas intact, or a full verification packet — manifest, claims, validation results, reviewer log — the audit trail a client or auditor can sign off on.

One honest limitation

On very dense drawings, AI vision can occasionally misread a note or legend entry as a symbol. That's exactly why every claim carries evidence back to its source region, and why nothing becomes a final quote line without a human looking at it. Trust, but verify — that's the whole product.