Getting started
Automation Backend
Build one backend that runs every automation in your company, with webhooks, scheduled jobs, an event ledger, and background workers instead of a pile of Zapier and n8n scenarios.
Coding harnesses and agent frameworks are transformational, and they earn the attention they get. They also make it easy to miss the low-hanging fruit. The instinct now is to reach for a model, and most of what a company should automate is not a model problem. It is an orchestration problem.
A company is a collection of systems, processes, people, and information that has to be orchestrated. Build one layer where that happens, and it becomes economically viable to build custom software for work that could never justify the effort before.
Datalumina OS
Datalumina OS is a self-hosted, event-driven backend that takes every webhook and scheduled job in the company, records each one as an event in a durable ledger, and runs the work in background workers that act on the systems the business already uses.
Every example, diagram, and tool choice on the following pages comes from that system. Swap any box for the equivalent in your own stack and the architecture still holds.
How this started
Over the past six months Dave replaced almost every Zapier, Make, and n8n automation at Datalumina with a single backend.
For years the low-code tools were simply easier. Even as an engineer, you reached for Zapier because clicking two nodes together beat writing, hosting, and maintaining your own script. That trade has flipped. An AI coding agent cannot read a canvas, test it, or change it. Code it can read, test, and change. That turned into a company rule. Everything is code first, so it can be AI first.
The starting point was the GenAI Launchpad, the event-driven template Dave built for AI applications. Datalumina OS applies the same pattern to a whole company instead of one product.
What the AI does
AI shows up in two places. It is used heavily to build the system, and it runs as one optional step inside a workflow when a task needs a model. Everything else is deterministic, on purpose, because that is what makes it reliable.
So calling a system like this an AI operating system would be wrong. It is a workflow orchestration engine, and most of it is ordinary backend engineering.
What makes it work is that it is event driven. Nothing calls another system directly and waits for the answer. Something happens, the system writes it down as an event, and the work to handle that event is picked up separately. That record is the ledger, where every piece of work is persisted, tracked, retried, and reported on.