Takeoff Engine v2 - auto drawing takeoff¶
| Stage | Estimated |
|---|---|
| Status | Deferred |
| Design status | In progress |
| Estimate | 3w (XL) |
| Confidence | Medium |
| Linear | PIN-694 ↗ PIN-440 ↗ PIN-441 ↗ |
| Linear status | Todo · +2 backlog |
| Cycle | C17 |
| Design | Design ↗ |
| Linear epic | Drawings, Documents & Takeoff |
| Module | Takeoff Engine ↗ |
Priority: High
Scope¶
Takeoff Engine v2: the September accuracy push on top of the shipped AI auto-takeoff base pipeline, plus a two-surface UI — a compact Summary Dashboard of system-level totals on the Estimate page, and a full-screen Takeoff Engine popup with itemized Quantities and AI-training Concordance tabs for two-monitor review — and a move from freetext material entry to an admin-managed material/system-tag dropdown. Progress is data- and model-strategy-gated (clean drawing corpus, self-hosted model track), so it matures iteratively into September rather than shipping as one fixed-date release. (Source: PIN-694, PIN-440, PIN-441.)
Acceptance criteria¶
- Summary Dashboard panel on the Estimate page shows system-level totals only (e.g. LF of pipe by type, lbs of duct by pressure class), not a per-size breakdown
- "Open Takeoff Engine" button opens a full-screen popup, usable alongside the PDF drawing viewer on a second monitor
- Popup's Quantities tab: full itemized breakdown by system (duct by size/gauge, pipe by size/material, valves, hangers, liner sq ft), collapsible per system
- Popup's Concordance tab surfaces AI-uncertain items for estimator review/training
- Material field becomes a dropdown sourced from an admin-editable materials table (per system tag), instead of freetext
- AI-extracted material text fuzzy-matches to the closest catalog entry; unmatched materials are flagged for review
- Scope matures through accuracy passes gated on a clean training-drawing corpus and the model-hosting decision, not a single fixed ship date
Conor's comments¶
This requires a lot of design discussion. This is a tough build and has to be PDF focused. We need to build training fiedls and train ai on detection and estimation.
Notes¶
Auto-measure from drawings; estimated high per Ivan S (don't estimate low)
Open question for Conor¶
-