Case study
Internal toolBrand Knowledge Hub & AI Listing Generator
A local web app holding each brand's products, branding and images in one place, with an AI chat grounded on that brand to generate on-brief images.
FIG. 001 / ARCHITECTURE
How the system connects
Live model
Problem
Producing on-brand product images and listing content for several brands required re-explaining each brand's products and packaging details every time, with no single place holding product facts, branding assets and past generated images together for a model to draw on.
System
A locally-run Next.js/Prisma app organizes brands into products, branding elements and a tagged image library, all versioned. A Claude-powered chat is grounded on the active brand's data through tool-based retrieval, so a plain customer brief (text or image) is enough for it to identify the right product, decide what to generate and at what size, and trigger generation through the Magnific MCP server, saving results back into the brand's tagged image library.
FIG. 002 / HOW IT WORKS
How it works
An early, self-hosted take on the same problem UBA-BRAIN and the visuals pipeline later solved differently: instead of a markdown folder read by CLI tools, this is a single-operator web app where a brand's full knowledge base lives in a local database and a pre-grounded chat is the interface for generating on-brief images. It runs entirely locally with no cloud platform dependency, and its test/demo data uses a public example brand rather than any real client's.
Stack
What it runs on
Related systems
UBA Panel
Internal toolA read-only status dashboard that shows whether the UBA-BRAIN system — clients, skills, health checks — is in good shape, from any device on the network.
UBA-BRAIN
Internal toolA shared markdown knowledge base and rule system that lets any AI CLI on any machine run the same client work with the same rules, memory and skills.
TruckPulse
Internal toolAutomated load-profitability and dispatcher-performance monitoring for a trucking company, built from Gmail broker PDFs and Telegram driver-rate messages.
Want a system like this for your team?
We start by mapping one workflow on a 30-minute call.