Case study

WabiRoom. Redesign the room, preserve its structure.

WabiRoom lets people explore a room's possibilities through virtual staging and redesign. Our AWS architecture brings guided capture, image generation and private delivery into one workflow, with quality and structural fidelity defined as explicit acceptance criteria.

Practice AWS Cloud · DevOps Engagement Platform architecture · virtual staging · mobile apps
Variations per photo
3
The generation workflow is designed to return three alternatives.
Delivery target
<5 min
For 95% of photos, including all three variations; not a measured result.
Quality target
80%
First-generation approval on a 50-room benchmark.
Availability target
99.9%
Monthly objective; measured post-launch results are not reported here.
The brief

A convincing image still has to be the same room.

Homeowners and property professionals need redesign images that retain the room's walls, windows and viewpoint. A visually attractive output that changes the underlying space does not solve the staging problem.

The platform also needs to accept varied phone photos, handle bursts of image-generation work and keep originals and results private. Arabic and English web and mobile journeys are part of that scope.

The approach

Make image quality a gate in the workflow.

Define the benchmark before scaling

A set of fifty real room photos provides the quality benchmark. First-generation approval and preservation of room structure are acceptance criteria, so quality can be evaluated consistently.

Prepare and check the input

Guided capture and upload processing prepare the photo, remove location metadata and check that the image is appropriate for the workflow before generation begins.

Separate generation from interactive traffic

SQS queues feed workers on EKS Fargate, with Amazon Bedrock providing room analysis and mask-controlled image generation. Queue-based scaling lets generation demand vary independently of the web experience.

Deliver to a private gallery

Encrypted image storage and signed CloudFront URLs control access to the results. Quality and moderation checks precede delivery notifications, with cost per photo tracked alongside job timing.

Design objectives

Acceptance criteria with room for measured results.

The architecture covers web and mobile capture, three generated variations, output checks and private gallery delivery. It defines a repeatable operating model with infrastructure code, deployment automation and monitoring.

The targets are delivery of three variations within five minutes for 95% of photos, at least 80% first-generation approval on the benchmark, and preservation of walls, windows and camera angle in every approved output. Post-launch measurements are not yet reported here, so these figures remain acceptance targets rather than achieved results.

The brief

The person you brief is the person who solves it.

One conversation · 30 minutes · no monologue