Launch HN: Rudus (YC P26) – AI for concrete contractors

Concrete subcontractors have one of construction's most stubborn bottlenecks: estimating. Before a firm can even bid on a job, a senior estimator has to open structural PDFs, manually trace every footing, wall, and column, then build out a spreadsheet with hundreds of line items covering concrete volume, formwork, and rebar down to bar size and lap splice length. That process can eat weeks, sometimes months, and most firms only have a handful of people who know how to do it. Rishi and Sahil, the founders behind Rudus, say that's the real ceiling on growth for these companies — not demand, but estimator bandwidth.

The idea for Rudus came out of a construction management class Sahil took, where he noticed the estimation workflow looked frozen in time. The pair spent time cold-calling firms, visiting job sites, and literally showing up with donuts to talk to estimators. What they heard consistently was that slow bidding is the industry's biggest constraint, and that every software product they'd tried before had failed them. The reason, according to the founders, wasn't lack of ambition but lack of trust — estimators are putting their name on bids worth millions, sometimes billions, of dollars, and they won't hand that over to something that behaves like a black box.

So Rudus is built to sit inside the existing workflow rather than replace it. Upload a set of structural PDFs and the system auto-classifies each sheet — foundation plans, section details, footing schedules — then uses computer vision to detect concrete elements and trace cross-references across the drawing set, catching details that tools relying on a single sheet would miss. Each detected element gets expanded into full line items for concrete, formwork, and rebar. The founders say a typical foundation package that used to boil down to a handful of assemblies now generates 80 to 120 priced line items automatically, which the estimator then reviews and can override.

The bet here is narrow by design. General AI takeoff tools built for general contractors treat concrete as a single checkbox, and generic vision models struggle with concrete drawings because the sheet conventions differ so much from other trades. Rudus instead trains proprietary models directly on customer takeoff data, with every correction an estimator makes feeding back into the system to sharpen accuracy for that specific client over time. The founders spent more than 100 hours sitting with concrete estimators and doing takeoffs themselves before building anything, which is a more grounded approach than most AI startups bother with before shipping a product.

The pitch, ultimately, is about defensibility rather than automation for its own sake. Rather than trying to fully automate a task where current AI output still needs to be redone by hand, Rudus keeps the estimator accepting, overriding, and editing every step. The founders argue that's the difference between a tool that speeds up work someone can stand behind in a bid room, and one that just produces numbers nobody trusts enough to use.