CARS Control: an AI dashboard on top of an ERP that could not be touched
Six areas in one dashboard (sales, stock, purchasing suggestions, credit, finance and supply) that read from the existing ERP without replacing or modifying it.
- Sector
- Automotive
- Role
- CTO
- My part
- Architecture and the integration with the ERP.
- Period
- May 2026 to present
- Status
- In production
- Team
- Blanco Brand team
System blueprint
The flow, in text
- The ERP stays as it is: neither replaced nor modified. It exposes its data through SOAP and SQL Server.
- CARS Control reads that data and keeps it in a read cache in MongoDB, so the ERP is not loaded.
- The dashboard and the AI module work on the cache.
Platform: NestJS 11 · React 19 · LLM through OpenRouter · Diagram. Not live data.
Product in use
Problem and constraints
The company’s commercial and operational information lives in Manager, its ERP. The ERP does its record-keeping job well, but answering a management question (what sold, what is running low, what to buy) meant exporting reports and cross-referencing them by hand.
Replacing the ERP was not an option, and neither was modifying it.
Constraints
- The ERP could be neither replaced nor modified.
- Dashboard queries must not load the system the operation depends on.
Decisions and architecture
I chose to build on top, without touching the ERP. CARS Control reads the data by two routes, a SOAP client and SQL Server queries, and presents it as a dashboard.
I put a read cache in MongoDB between the dashboard and the ERP. The dashboard queries the cache, not the ERP, so queries are fast and do not load the system the operation depends on.
I added an AI module that answers questions about that same data, with language models connected through OpenRouter.
The API is NestJS 11, with authentication on AWS Cognito and monitoring with Sentry. The dashboard is React 19 with Vite 7, Tailwind CSS v4 and shadcn components.
Integrating with someone else’s system is where things break without warning, so the project has 149 test files on Vitest.
Results
CARS Control is in production and covers sales, stock, purchasing suggestions, credit, finance and supply.
- 6areas in one dashboard: sales, stock, purchasing suggestions, credit, finance and supply.
- 149test files across the API and the dashboard.
- 6months of continuous development, May to November 2026.
Limits
- CARS Control reads from the ERP: it neither replaces nor modifies it.
- This case does not publish evaluation, latency or cost metrics for the AI module.
Do you have a system that cannot be touched?
Tell me the problemWant the full career history?
View the CV