VARAi: the operating system for agentic commerce

1. P-01AI Designer & Material StudioLive AI design chat · AI that acts, not just answers · 3D materials · Render automationChat → live 3DDesign mode changes materials, lighting and exposure in the scene; Apply or Reject each change in the chat
2. P-02Lens JourneyDeterministic rules engine · Prescription upload · Catalogue sync · Multi-platform commerce1,500+ testsLens-journey suites, Sep 2026 (recorded runs of 1,431 and 1,772)
3. P-03Virtual try-onBrowser AR · 3D rendering · Storefront integration · Production reliability38.3 s → 57 msOne documented gallery-query benchmark
4. P-04AI search and recommendationsAI search · Recommendations · Redis caching · Magento integration~2 s → ~50 msRepeat queries served from the Redis cache, skipping AI routing and the store's API
5. P-05Store analysis and catalogue syncCatalogue sync · Data mapping · Shopify · Magento · Human approval43 products · 1,605 variantsLive run on a heritage British eyewear store; 6 accessories excluded and named

Yes, and make it pay its way. These run for paying customers.

AI search

A fast LLM for intent, Redis for speed, live store data for prices and stock. Replaced the hosted search service, with a 107-query side-by-side test against it. Instant results as you type.

~2 s → ~50 ms cached

Recommendations

A "you may also like" bar of four to eight products, built on the same engine.

AI Designer

A chat that edits a live 3D scene. QA mode gives tips; Design mode makes the changes, with Apply buttons in the conversation.

Store analysis

Samples a new client's catalogue, finds which fields really hold colour, size and fit, and proposes mappings with measured confidence for a person to approve.

1,605 variants / test run

Where a wrong answer costs money or trust, I don't let a model guess.

Lens Journey

A ten-step prescription-lens configurator that makes an optometrist's choices. A rules engine with JSON rulesets of 650–5,100 lines and thousands of paths. The lens logic uses no AI by design; AI only reads an uploaded prescription for the shopper to check.

1,500+ tests

Sync gates

Store analysis proposes; nothing changes until a person confirms. Catalogues with up to 50 variants per SKU sync the same way every time.

50 variants / SKU

AgentOps

My audit harness: a validator, not the model, decides what counts. Read it.

146 / 37

Multi-region

US, EU and India, each with its own data, for data-residency law.

3 regions

Live migration

21 enterprise tenants moved from the EU to India behind row-count safety gates.

0 data lost

Staging gate

A production mirror with a demo store and end-to-end tests. Every change passes it before production, for every client instance.

100% of releases

DevOps & security

GCP kept secure, fast and cost-effective: security and billing audits, infrastructure builds, CI/CD.

74 services

Scale

Client brands onboarding fleets of stores, plus a ten-app enterprise programme.

~150 stores

All day. I work in Claude Code and Codex for most of every working day, and I build the guard rails that make that safe for a team.

Agentic git

Gates, rules and review processes that keep thousands of agent-generated pull requests shippable across a multi-developer team.

Claude Code gates

Agents must pass staging before anything reaches production.

24/7 audits

AgentOps runs on a remote VM with its own dashboard, checking the 221 architecture decisions currently on record against the code.

221 monitored

Virtual try-on

Frames on a moving face in live video. The first build on MediaPipe and three.js wasn't good enough, so we replaced it with Visage's SDK and shipped.

Material Studio

Custom materials (colour, lighting, HDR), saved as templates, edited per mesh with a live 3D view.

3D automation

Scripts that render thousands of identically framed product thumbnails and turntable videos from each client's 3D models.

1000s / client

• —Prototype on the vendor SDK before building your own. Our first try-on cost weeks before we replaced it.
• —Build the staging gate on day one. Features that passed tests but did nothing in production taught us that.