Aveline
Our in-house multi-agent concierge platform for semi-luxury boutiques — WhatsApp inquiries, customer memory, visual sourcing and owner-approved checkout.
Overview & Strategy
Aveline is not a client commission. It is a product we designed, built and shipped ourselves, and it is live today at aveline.gravora.dev. When a boutique's customer messages in on WhatsApp or Instagram, three specialist agents answer together. Ava reads the message and recalls preferences, occasions and past purchases from a pgvector semantic memory. Elle studies a reference photo, searches the shelf and composes outfits that suit the moment. Lina checks the margin, drafts the deposit and plans the courier. Anything that crosses an owner-set threshold stops and waits: every high-impact action is approved by a person before it ever reaches the customer. Underneath sits an ASP.NET Core modular monolith, a Python and LangGraph agent service, a Flutter app for floor associates, a React dashboard for owners, PostgreSQL with pgvector, Redis, and an Azure Container Apps deployment rolled out by GitHub Actions.
Screens From The Live Product
What It Does
- WhatsApp and Instagram inquiries land in one shared workspace
- Ava: customer memory on PostgreSQL with pgvector semantic recall
- Elle: photo-to-shelf visual matching and outfit composition
- Lina: margin checks, deposit drafts and courier planning
- Human-in-the-loop approval queue before any threshold is crossed
- Flutter floor-associate app plus a React owner dashboard, role-gated by Clerk
- Blossom credits meter every agentic action