How an AI call centre works
We do not run a blog and we do not publish case studies we do not have. Instead we show the working documents of the system.
Team training — 10 topics, 50 minutes
The same programme that lives inside the product. It is in Ukrainian only for now: a machine translation would ruin exactly what it is for.
What we can show you
- 01
Architecture
Why a modular monolith, and why Postgres is the queue, the search and the vector store at once.
- 02
Latency
What makes up the seconds between “the customer stopped talking” and the first sound of the answer.
- 03
Cost per call
A calculation at real provider prices and a comparison of two voice pipelines.
- 04
Test scenarios
How conversations are replayed by a model before every release.
- 05
Architecture decisions
Twelve recorded decisions — what was chosen and why.
- 06
Operations
How the system is deployed, updated and restored.
Want to look
inside?
Not slides — the working system with its logs open.