Code-compliance on autopilot
This may be counter-intuitive, but the parallels between self-driving cars and code-compliance software run deep: both have been holy grails for over half a century before AI forced us to see them as live options again; many people have doubted the possibility of its success; and yet, the societal benefits of both technologies is word-changing, so technologists have strived to make headway nonetheless.
Both AV and code compliance software start with the same first task: see the world with extreme precision and reliability.
That reliability gets stress-tested relentlessly. AV teams talk about long-tail events, like a mattress on the freeway, or two cyclists in one lane. Architectural plans have their own long tail: over a hundred thousand pages of regulations that have to apply to everything from a football stadium to a 1940s remodel. You cannot script all these cases. You train big models on big, diverse data, then you keep feeding them edge cases just like Waymo and Cruise do with rare driving scenes.
We see AV as among the most brilliant technologies of our time - our jaws dropped when we first rode in a Waymo. So we spoke to ML Engineers at Waymo and Cruise to understand how this technology succeeded. A core lesson Waymo and Cruise ML engineers harp on is that evaluation has to be a core focus - it unlocks step change improvements in your technology.
We do the same. Half of our engineering time goes to tests, especially tricky synthetic plans, and dashboards that track even a one-inch drift in segmenting a sink.
Don't judge our competitors by their demos. Instead ask them for the size of their evals and ask to see the architectural plans in their evals. If their architectural plans in their evals look nothing like yours, they're not building for you.
This hill is steep, and history says most teams will fail. Construction-tech writer Brian Potter traced fifty years of failed code-checking projects, and one lesson is clear from history: shallow rule scripts break the moment reality shows up.
This problem is hard, and yet the stakes could not be higher: faster compliance checks lower the cost of building, and can secure a future of abundant housing, safer hospitals, and lower costs for everyone.
If this work excites you — if you love computer vision, permitting, or just shipping hard things — reach out to us at founders at planweave dot ai. The road from blueprint to building is long. We're excited to be driving it.