Census

Inspiration

US office vacancy reached a record \(21\%\) in early 2026, as per Moody's, and Houston's is about \(25\%\). The same cities are short of affordable homes. Los Angeles and Texas changed their laws in 2025 and 2026 so that offices can convert to housing without re-zoning. The remaining obstacle is the building. Checking whether one floor can become apartments takes weeks of architects, estimators and code consultants, so most floors never end up getting checked.

What it does

Census screens a vacant office floor in seconds, allows a developer to do an in-depth online survey, and traverse the most optimal layout for low-cost housing.

For a chosen floor, Census:

Measures the floor plate, core and ceiling height from the interior scan and aerial outline
Lays out income-restricted apartments
Lets the user walk through every home in 3-D, with sunlight and shadows from the real neighbouring towers
Will check about compliance rules across entitlement, life safety, habitability, accessibility, metering and affordable-housing requirements

How we built it

CoStar: Vacant floors, asking rents, building facts and nearby apartment rents
Matterport: Interior scans used to measure the floor plate, core and ceiling height
Vexcel: Building outlines, the surrounding city in 3-D and the heights that cast shadows

The back end is Python, with separate modules for the floor plate, layout, sun, rules, costs and feasibility. We also built a column study that marks \(22\) probable structural columns in the scan of The Oppenheimer Tower, Suite 1600.

Challenges we ran into

Registering the interior scan to the aerial outline, solved with a rotation search
Generating layouts on curved and irregular floor plates
Placing doors, bathrooms and kitchens workably in narrow homes
Retrieving the load-bearing beams and wall-less floor plan

Accomplishments that we're proud of

Census works on real buildings: every layout comes from a measured scan and aerial survey, not a mock-up
It gives honest answers, including that Wells Fargo Plaza's deep floors fail the natural-light rule
You can walk through every generated home and watch real neighbouring towers cast shadows across it
The same pipeline ran on a second tower in a different city and state without changes to the method

What we learned

Floor depth and compliance constraints decide most conversions, as zoning is no longer the main barrier in either city
For affordable housing, the deciding figure is the funding gap per home, not profit
The share of a tower that converts determines which state protections apply

What's next for Census

Screen every building with a vacant floor and a scan, ranked by homes, cost and funding gap
Add AI agents to research local costs, review code text and iterate layouts
Extend the method to hotels and retail