Case Study
PureIce — AI-Driven Diamond Curation
A shopping experience that uses AI to help you find the best diamond for your engagement ring.
Pure Ice turns certificate data into a single, explainable beauty score for every diamond, then uses those scores to put the three best stones for a shopper's shape and budget in front of them first.
- Applied AI
- Curation
- Commerce
- Recommendation
- 10
- Shape-specific scoring models
- 1M+
- Diamonds per supplier feed
- 0–100
- Beauty Index, with 7 subscores
- Top 3
- Picks per shape and budget
The problem
A certificate describes a diamond. It doesn't tell you if it's beautiful.
Diamonds are sold on the 4Cs, but beauty comes from proportions: how the table, depth, crown and pavilion angles work together to return light. Two stones with identical grades can look completely different face-up.
Online, that becomes a wall of near-identical listings. Shoppers either learn gemology or trust a grade that doesn't answer the question they actually have: which of these will look best?
The Beauty Index
Scoring beauty, shape by shape
I built a scoring engine that reads every measurement on a diamond's certificate and produces a single Beauty Index from 0 to 100. A great oval doesn't follow the rules of a great round, so each of the ten shapes has its own model.
- Each measurement is compared with the ideal and acceptable ranges for that shape, then weighted by how much it changes what the eye sees.
- Rare qualities earn credit, such as Hearts & Arrows potential in a precisely cut round. Visible flaws cost points, such as a bow-tie shadow across an oval.
- Results break down into seven subscores a shopper can read: Light Performance, Fire, Scintillation, Shape & Proportion, Visual Size, Clarity Impact and Finish Quality.
- Each score lands in a tier, Exceptional, Elite, Premium or Good, so where a stone ranks is clear at a glance.
Curation
From a million listings to three
Scores are only useful if they change what a shopper sees. Every scored diamond is indexed by shape and budget and ranked by Beauty Index, with value breaking ties. The shopper picks a shape and a price range and sees the top three diamonds in that category, each with its full beauty report.
The indexes rebuild every time inventory refreshes. When a stone sells or a better one arrives, the top three change on their own, with no manual merchandising.
Engineering
Built for messy supplier data at scale
- A batch pipeline ingests a supplier feed of over a million diamonds, normalizing every field, quarantining invalid rows, removing duplicates and auditing each run, so bad data is never silently dropped.
- A normalizer turns inconsistent free-text supplier values into structured data and falls back gracefully when a value can't be parsed, so one malformed field never blocks a score.
- All scoring logic lives in a reusable class library behind a thin API. A concurrent batch worker scores the inventory with configurable parallelism and stores the full breakdown for every stone.
- The customer storefront, checkout and admin portal run on the same platform, backed by a single relational schema.
Outcome
Three explained choices instead of endless filtering
Pure Ice replaces "trust the grade" with one consistent, transparent beauty standard applied to every stone in the catalogue. Shoppers start with the best diamonds they can buy at their budget and can see exactly why each one made the list.
Built with
- .NET
- ASP.NET Core
- Blazor
- C#
- MySQL
- EF Core
- Docker
- n8n
- Stripe