Notes & writing
Sharp notes, not filler.
Things we have learned building AI, software, and infrastructure. Written for people who build, not for the algorithm.
RSS feed- 4 min
How a local model keeps a 67-language site translated
Translation APIs price small teams out of being multilingual, and agencies price everyone out of staying multilingual. A local model and a ledger of what changed solve both, for the cost of electricity.
AIOn-premisei18nEngineering - 5 min
Why we run AI on our own machines
On-premise AI is not nostalgia for owning hardware. It is a practical answer to three real problems, data control, cost, and dependence, and it is more achievable in 2026 than most teams assume.
AIOn-premiseInfrastructure