We built it with estimating teams of commercial sheet-metal contractors in Chicago, around the takeoff they already did by hand in Bluebeam and Excel.
Before a sheet-metal contractor can bid on a building’s ductwork, an estimator has to measure all of it. That means opening each mechanical plan in Bluebeam, tracing every duct run by hand, reading its size off a label, counting every elbow, take-off and register, and typing the totals into a spreadsheet that turns feet of duct into pounds of steel. A single bid set can take days, and a contractor bids on far more jobs than it wins.
The cost of a mistake is not abstract. A misread size or a run counted twice moves the weight, and the weight is the price. One question in the demo, whether 70 lightly drawn runs are existing duct that stays, decides about $170,000 of the bid on its own.
A mechanical drawing is not a diagram of ducts. It is lines: walls, grid, ceilings, piping, text and duct all drawn in the same strokes, often with labels drawn as lines too, so there is no text to read. New duct is drawn heavy and existing duct light. Scale notes are sometimes wrong, enlarged plans carry their own scale, and the same duct can appear on two sheets.
A tool that guesses is worse than none, because the estimator then has to check everything it did. So the problem is not only reading the drawing. It is knowing which of its own readings it can stand behind, and saying plainly which ones it cannot.
Rian Blue reads each sheet in stages, and shows every stage live as it happens, so the estimator watches the reading rather than waiting on it:
What comes out is not a report. Every duct the tool found becomes a line on the estimator’s own Bluebeam-style tool, shaded to the width of its size so a wrong size shows as a band that does not fit the walls. Everything stays editable, and a line the estimator changes is theirs: no later answer or re-reading overwrites it.
On the demo sheet, the whole reading took about 12 seconds. From 1,761 duct-style lines it paired 993 walls into 97 runs, read 116 size labels, confirmed the drawing’s 1/8" scale against 44 measured duct widths, and found the fittings and 14 air devices. The AI cross-check agreed with three of three labels it was sent.
Review then asked two questions instead of asking the estimator to check 97 runs. Each comes with the lines it is about lit on the drawing, a suggested answer, and how much of the bid depends on it, so the expensive questions are the ones answered with care. One key gives the suggested answer, one skips, one goes back.
The finished takeoff fills the estimator’s own workbook: one tab per duct type, one row per size, every run in its own cell, with their formulas for perimeter, lining, waste and pounds per square foot left live. The drawings export as a PDF whose marks are ordinary markups that open in Bluebeam.
Single-line duct drawings, oval duct and duct shown only in sections are not measured yet. On scanned drawings, lengths are a few percent less exact than on a clean PDF and fittings are not counted. Duct that appears on both an overall plan and an enlarged plan is not merged, and the weights come from a standard gauge table rather than each contractor’s own until they type theirs in. The tool says each of these plainly where it applies.
Buyers on a promotional-products franchise network knew what they wanted but not what it was called, and keyword search across more than 100,000 products could not help them. An AI search now finds products from a photo, a phrase, or both, on the head-office site and every franchise storefront.
A marketing team running accounts on six platforms was writing every post six times. We built one workspace where a post is written once, rewritten for each platform from the team’s own past posts, scheduled, and measured in the same place.
Patients with depression see their psychiatrist for minutes at a time, and the weeks in between go unrecorded. Sogon-Sogon is a voice-first AI journal that turns a patient’s day into a diary entry for them and a report their clinician can read.
