The problem
Capturing a space in 3D usually forces a tradeoff. Professional rigs — laser scanners, camera arrays, dedicated depth hardware — are expensive, heavy, and slow to operate. The consumer tools that run on a phone tend to produce rough, blocky geometry that looks nothing like the real room and can't be trusted for measurements.
Getting output that is both photorealistic and dimensionally accurate, from hardware people already own, is genuinely hard. That gap is what Spatia set out to close.
What Spatia does
Spatia uses the LiDAR sensor and cameras already built into modern iPhones to capture a space, then reconstructs it as a 3D Gaussian Splat — a state-of-the-art, photorealistic 3D representation you can fly through in a browser — alongside a clean, measured floor plan.
You walk through a room for a couple of minutes; you get back something you can explore, measure, and share. No rig, no markers, no special hardware — just the phone in your pocket.
How it works
Capture
You point the phone and walk. Spatia reads depth from the LiDAR sensor and tracks the camera's motion in real time, with live on-screen guidance that shows which parts of the room are already well-covered and where to aim next — so first-time users get a good scan on the first try. As it records, the app intelligently selects the highest-quality frames rather than blindly keeping everything.
Live capture guidance running on-device during a room scan.
Reconstruct
The captured images and depth are handed to a custom training pipeline running on on-demand cloud GPUs, which reconstructs the scene as a photorealistic 3D Gaussian Splat. Because the capture carries real-world depth and scale, the output is metrically accurate — distances in the model correspond to distances in the real room.
Explore
The finished splat streams to a web viewer that renders it interactively in the browser — nothing to install. You can orbit the space, look around from any angle, and take measurements straight off the model, as shown at the top of this page.
Technical highlights
Spatia is a full vertical slice across four hard domains that rarely live in one project:
Good scans on the first try
Real-time capture fuses the phone's LiDAR depth sensor with camera-motion tracking, and live guidance coaches non-experts through full coverage — while the app selects the highest-quality frames as it goes.
Photorealistic and to scale
A custom training pipeline turns the captured images and depth into a photorealistic Gaussian Splat. Because the capture carries real depth and scale, the result is metrically accurate, not just good-looking.
Predicting quality before you compute it
Spatia can estimate the best quality a scan could ever achieve before any expensive reconstruction runs — grading the raw capture itself. That turns capture from a black box into a measurable, improvable step and avoids spending compute on data that could never produce a good result.
From phone to browser
Reconstruction runs on on-demand cloud GPUs; results are served through a web front-end that renders the interactive 3D splat in real time, backed by an API and a relational database.
My role
I built Spatia end-to-end: the iOS capture app, the 3D reconstruction and machine-learning pipeline, the cloud compute that runs it, and the web viewer that serves the results. It's the project where my mechanical-engineering instinct for real-world measurement meets the data-science and systems work I've been growing into — and it's an active, independent project I'm still pushing forward in 2026.