Data engineering — and why data between systems quietly goes wrong
I work on data in motion: parsing and normalising what arrives from other systems, moving it, reconciling what does not match, and the analytics built on top. Java, Python, Go. Nineteen years of backend and distributed systems.
Most of what I publish is about the quiet failures — duplicates, gaps, stale values, order and reprocessing. They produce no errors in the log and surface weeks later as a wrong number.
Where those numbers feed invoices, settlements or stock, the failure stops being a technical detail and becomes money — charged twice, never invoiced, or never counted at all.
In the Odoo Community Association I proposed a fix that keeps queued jobs from losing the name of the method they were asked to run (OCA/queue#998).
Earlier, as an independent contractor, I worked directly for telecom companies in Slovenia — Iskratel and RC IKT — remotely, with acceptance testing on the client's own equipment.
When I publish a number, the code that produced it and the data it ran on go out with it — a disagreement should be with a rule you can read, not a figure you have to believe.