The technology

How Smores works.

A full-stack system for capturing and streaming real-world 3D in real time — hardware, AI, and apps, all built in-house.

System overview

One pipeline, end to end.

STEP 01

Capture rig

A portable multi-camera ring records the real scene from every side.

STEP 02

Reconstruction & streaming

Our AI fuses the feeds into a live 3D scene and streams it in real time.

STEP 03

Viewer apps

Explore the scene on phone, web, or in a VR headset.

Capture hardware

Capture rig — our own hardware.

We design and build our own multi-camera capture rig — a portable ring that sets up in minutes with no markers, wearables, or room modifications. It runs on 8 cameras today, with a path to 2–3 as our models improve.

Custom multi-camera hardware Portable, fast setup Raspberry Pi firmware
The Smores capture unit — wireless and battery-powered
The Smores capture unit — wireless and battery-powered
Fixed 45° mount, or a flat cover for walls and corners
Fixed 45° mount, or a flat cover for walls and corners
Sticks on in seconds with a single command strip
Sticks on in seconds with a single command strip
Embossed Smores branding on the housing
Embossed Smores branding on the housing
Wide field-of-view camera and status LED
Wide field-of-view camera and status LED

Reconstruction & streaming

Reconstruction & streaming engine.

Our cloud engine fuses the camera feeds into a live, explorable 3D scene and streams it in real time. Reconstructing a full 3D scene from just a few views is the hard AI problem — and it's the core of what we do.

Python + PyTorch TensorRT acceleration Real-time streaming Gaussian-splatting-based reconstruction
[ videos/reconstruction.mp4 ]

Viewer apps

View it anywhere.

The same capture plays across devices — a phone and web app for everyday access, and a native VR app for full immersion.

Mobile & web

Flutter app (iOS today) and browser viewing.

VR

Native Meta Quest app built on OpenXR + Vulkan.

Same capture, any device

One scene, viewed however you like.

The moat

The data flywheel.

Every capture generates rare, real-world 3D data — proprietary, and hard to scrape or buy. That data trains our models to do more with fewer cameras and unlocks new capabilities. More captures → more data → better models → lower cost and wider reach.

01

More captures

02

More 3D data

04

Fewer cameras + new uses

03

Better AI models

Reference

Tech stack summary.

Mobile / web app
Flutter (iOS), web viewer
VR app
OpenXR + Vulkan (Meta Quest 3S)
Server / models
Python, PyTorch, TensorRT
Reconstruction
Gaussian-splatting-based
Capture firmware
Native Raspberry Pi
Hardware
Custom in-house multi-camera rig

Curious about the technical details, partnering, or joining us?

Get in touch Request a demo We're hiring