DeLightfully Made 818

The problem

A craft retail business was running products, orders and materials through spreadsheets and messages. Every change required the owner personally, which made the business unable to grow past the owner’s available hours.

The brief was a storefront. The actual constraint was coordination overhead.

What shipped

A mobile-first storefront in Next.js and TypeScript, and — the part that mattered more — an admin system that lets non-technical staff manage products, orders and materials without a developer in the loop.

That admin layer is what removed roughly 20 hours a week of manual coordination. The storefront was the visible deliverable; the admin tooling was the one that changed how the business ran.

Inventory and payments

The backend runs on Supabase with real-time inventory sync, so stock and order records stay consistent without anyone reconciling them by hand. Payment processing is handled securely rather than rolled by hand — the correct call for a small retail client where a mistake means real money and real liability.

Two agents grounded in live business data

Two AI agents run in production, grounded via RAG in the business’s live product and order data. They handle routine customer enquiries, which cut the owner’s manual response load.

The grounding is the important detail. An ungrounded assistant on a retail site invents stock levels and prices — confidently, and in a way that costs the client a sale or a refund. Retrieving against live records means the agent answers from what is actually true right now.

From answering to acting

The agent layer was then extended to take multi-step action on business data: inventory updates, order processing, payment tracking. That is the step from a chatbot to something that does work, and it is also where the failure modes get expensive — which is why it came after the read-only version was proven in production, not before.