Vision AIAug 2026 · Beauty Decoded hackathon · Team of 2

SHADE

Your true colours, in writing.

A vision-model colour analysis. It names your colour season with a confidence score and plain-language reasoning, then scores real retailer products on shade match and on your ingredient avoid-list.

Open the live app (opens in a new tab)Skills it shows

my-colour-guide.lovable.app

The live product, built and shipped in one day.

RoleDesigned and built the product

TeamTeam of 2

WhenAugust 2026, one day

Built withVision model, Lovable

Skills it shows

Applied vision AI

Reads face, wrist, eye and hair photos and returns a colour season with a confidence score and its reasoning.

Confidence thresholds

Below 70% confidence it asks for a retake in daylight instead of guessing.

Ranking design

Two separate scores on real retailer catalogues: shade match, and fit with your ingredient avoid-list.

Inclusive, private by design

Tested across the full range of skin tones, with any failure blocking release. Photos are processed, deleted and never used for training.

Problem, build, result

1Problem

A professional colour analysis costs £150–£500. Free apps name a season with no reasoning, and nothing you can use in a shop.

2What I built

A reading, a 30-shade palette made from it, and a filter that scores real products against both the palette and your avoid-list.

3Result

A live product that shows its working and states its confidence, built and shipped inside the hackathon day.

70%confidence floor before it answers

30shades in every palette

2separate scores per product

1 dayfrom brief to live product

In pictures

Select any image to see it larger.

The product in three columns: the reading, the palette and the filter, each with a short description. — Three parts: the reading, the palette, the filter.

A grid of thirty colour swatches beside the line “The interface is neutral so the only colour on screen is yours.” — A neutral interface, so the only colour on screen is yours.

Six product rules: show the working, confidence stated, bad photos refused, no medical claims, photos deleted, full depth range. — The product’s rules, written on the product itself.

Each reading arrives with its reasoning in plain language, and every ingredient flag cites its INCI name so it can be checked.

Tech stack

Tools and data

LovableVision modelRetailer cataloguesINCI ingredient names

Methods

Seasonal colour analysisConfidence thresholdTwo-axis scoringPrivacy by design

Next projectUsage Data PipelineA Node.js service that syncs vendor APIs and matches every record automatically.Read the case study