CLMS · a dealership runs through it

A used-car dealership is not one workflow, it is eleven people with eleven jobs that all touch the same vehicle. A car gets acquired, inspected, serviced, listed, marketed, test-driven, credit-checked, financed, insured, sold, and handed over. In most small dealerships those steps live in spreadsheets, group chats, and one person's memory.

The brief: build what DealerCenter does, scoped to how this dealership actually operates, so every actor works inside one system and a vehicle's full story exists in one place.

• Modeled the system around actors, not features: each of the 11 roles got its own dashboard, permissions, and queue, all reading from one vehicle timeline.
• GraphQL API over Node and MongoDB, so each role's dashboard fetches exactly the slice of a vehicle's state it needs.
• Credit and financing flow with staged applications: a lead can move from inquiry to credit application to contract without leaving the system.
• Inventory and servicing tied to the same vehicle record, so a car in the shop is visibly not on the lot.

Actor-first modeling. The earliest version was organized by feature (inventory, sales, CRM). It collapsed the moment two roles needed different views of the same data. Rebuilding around actors made permissions and screens fall out naturally.

GraphQL over REST. Eleven dashboards with overlapping but distinct data needs is the textbook case. One schema, per-role queries, no bloated endpoints.

One vehicle timeline. Every action any actor takes lands on the vehicle's history. That single decision is what made the system trustworthy to the people using it.

Deployed and used as the operating system of a real dealership workflow, from vehicle acquisition to final sale. It is the project that taught me how to model messy human processes without flattening them, a lesson I now apply to AI agent workflows daily.