AI · White-label

Miroir

Online, advice vanishes: abandoned carts, wrong choices. Miroir brings the expert advice of a beauty store back, at scale.

Miroir is a white-label conversational advisory infrastructure for beauty: an expert advisor that converses, educates on actives and perfumery, runs a cosmetic skin pre-diagnosis and steers to the real catalogue products. Where a generic chatbot improvises, Miroir keeps three product-grade promises: reliability, adaptability, governance. A real product, already in production on a real store, and testable live.

Try the live demo Live · testable demo

1,000+ real references plugged in
5 languages, incl. Moroccan darija
0 hallucination (catalogue-grounded)
3 layers of safety
~5× lower cost per message (cache)

TypeScript
Next.js
Supabase
pgvector
Claude API

Related expertise : SaaS cloud development →

Gallery

Modules

01

Never cites a product that doesn’t exist

Strict catalogue grounding: price, availability and references come from verified tools, never invented. Sourced and governed knowledge base (A→D reliability rating, expert validation on sensitive topics, versioned and traced ingestion). Without a safe source, the assistant stays qualitative and says so.
02

Speaks your customers’ language, darija included

A rare differentiator: authentic Moroccan darija, on top of French, Arabic, English and Spanish. Configurable persona (name, tone, formality, language level), adjustable length and format, from short chat to detailed advice.
03

Sells, without ever pushing

A sales assistant that adapts to catalogue, promotions, stock, channel and goals (featuring, cross-sell, trade-up). Recognizes the logged-in customer, retrieves address and order, proposes clickable products right away then refines.
04

One stack, all your brands

White-label and agnostic: plugged in via normalized contracts (Catalogue / Account) to Shopify, WooCommerce or another platform. Switchable AI model with no redeploy, cost per message reduced by roughly 5× through caching.
05

Governed to be reliable, not just chatty

Strict health doctrine: never a diagnosis, systematic referral to a professional. Three layers of safety, HMAC-signed exchanges, per-brand isolation (RLS). What the assistant knows and may say is controlled, traced, auditable.
06

Already in production, not a prototype

Proven on Maison d’Améthyste: a real 1,000+ reference catalogue plugged in, FR + darija advisor, clickable real-time recommendations, recognized customer. Fully decoupled, multi-tenant, streaming engine.

Under the hood

Under the hood: TypeScript / Next.js streaming (SSE) · Supabase — PostgreSQL, pgvector, multi-tenant RLS · Claude API — tool-use and prompt caching · plugged in via normalized Catalogue / Account contracts, e-commerce platform agnostic.

Live demo