4 engines. 3 layers we run ourselves.
26 people produce the output of 100 because the platform we started building in 2017 runs on an AI stack we operate ourselves, on our own iron. These are the engines behind it, and what each one is for.
Each layer has a job. And an engine that does it.
-
Layer 01
Anthropic, in the daily work.
Claude drafts listings, content and service replies inside the operation, not beside it. A named person owns every outcome that leaves the building.
Reasoning -
Layer 02
DGX in the building.
Our own iron in Patras for the work that should never leave it. Partner data stays on hardware we can walk up to and unplug.
On-prem -
Layer 03
Bedrock when it has to scale.
Cloud-scale models where elasticity wins the argument, on the AWS AI accelerator track. Peak season is a setting, not a rebuild.
Elastic
The fourth engine is Google, where we are an official Partner. Because it is also 1 of the 12 channels we sell on, it keeps a single page: how we run Google →
The stack is ours to operate. So the answer is ours to give.
A brand asking where its data sits, or what wrote a listing, gets an answer from the people who run the machine. That is the whole reason the stack is not rented.