Finch
Choosing the finishes on a new-construction home means committing thousands of dollars. Buyers get two ways to picture it, both flawed.
The big 3D visualization platforms can show the whole room, but they cost thousands a month and take months of per-floor-plan setup, so only the largest builders run them. A design showroom lets you hold the real samples — but one at a time, never composed together in the actual room.
Finch is the in-between: a buyer's exact selections rendered together, in real photos of the home they're buying, and live for a builder in days instead of months. I built it because I lived the problem.
It started with our own house
When my wife and I were buying new construction, the design center handed us a stack of finish samples and a price sheet and asked us to imagine the result. We couldn't, and like most buyers we second-guessed everything and held back on choices we'd probably have loved.
I built the first version of Finch for us: pull the builder's real options and prices, and use AI to render our actual selections in photos of the model home. Seeing it changed how we decided — even trying to stay disciplined, we added 40% more in upgrades once we could finally picture them. The buyers least likely to spend are exactly the ones who can't see the result, and we were proof. The whole premise is simple: buyers upgrade what they can see.
From our kitchen to a builder's pilot
Our real-estate agent saw what I'd built and asked to show it to her builder. That turned into a live pilot with a production home builder in Alabama, running on their real floor plan, real model-home photos, and real pricing.
How it works
Buyers open it on any device: no app, no account. They move through their finishes category by category, watch each selection render into the actual room, and see the running price update as they go. When they're done, a priced selection sheet exports for the sales team.
Builders manage their own options, pricing, and photos from an admin dashboard, and a new floor plan can go live in days instead of the months the incumbent 3D tools take.
The hard part is making it fast and real
The product only works if the renders look genuinely real and show up in seconds. Most of my engineering time goes into the generation pipeline: getting photorealistic images of specific materials in a specific room, reliably, fast enough that selecting feels live. It's an ongoing R&D loop: evaluating new image-generation models, tuning prompts surface by surface, and pre-caching the common combinations so the experience feels instant.
How I work now
By Finch, building this way had stopped being the experiment and become simply how I work. The open question isn't whether I can ship it; it's the frontier the product lives or dies on: photorealistic renders of specific materials in a specific room, generated fast enough that selecting feels live. That's a standing R&D loop of evaluating new image models and tuning them surface by surface, and it's where my time goes now.
It's the sharpest version of the pattern all three share: find a real problem I've lived, then design and build the answer myself.