Concept build: a demonstration project, not client work
A business intelligence dashboard for a multi-store retailer. It leads with the stores and products that changed, each with a plain-language explanation.
Sales against plan, by store
What changed: Store 07 is below plan because two top sellers were out of stock from Tuesday to Thursday.
Interface shown with illustrative data.
Take a retailer with a few dozen stores. Sales, stock and staff schedules live in separate systems. Each week, regional managers export spreadsheets, merge them by hand and argue about whose total is right. A simple question, like why one store's margin slipped, takes days. The answer arrives after the moment to act.
Aisleview starts with the plumbing: each night, Airflow copies every system into PostgreSQL. There, dbt models turn raw transactions into tested tables for margin, stock cover and labor cost. The Next.js dashboard opens on exceptions: stores and products outside their normal range. Each gets a short note built from the figures, and a click shows the query behind it. Store managers see only their own store.
The home screen ranks what moved by its effect on revenue or margin. No wall of charts.
Each figure is written once, in a tested dbt model. Charts and exports all read from it.
The language model only phrases numbers the warehouse already computed. It never does the math itself.
How a request moves through the system.
01Airflow pulls sales, stock and schedules each night
02Raw records land in PostgreSQL staging tables
03Tested dbt models compute margin, stock cover and labor
04A job compares each store with its own history
05Managers open the exceptions and drill into the rows
Next concept build
BrieflightProfessional servicesDescribe your version of the problem. We reply with questions and a proposed approach.