Case studies / six of eight, in full
AI systems in production
Each write-up follows the same spine: the problem, the architecture, the AI components actually used, what the platform runs today, and what changed as a result. Ommune built every one, and still operates every one. None of them is a product for sale.
Two further platforms are in production, alongside several internal tools, and are not written up here — their write-ups would say more about someone else's business than about ours.
What these have in common
Read together, the same three decisions show up in all six. The deterministic path is built first and handles the majority of cases. The model is tiered by cost, so the expensive call only sees what the cheap paths could not resolve. And every AI path has a defined safe result when it fails, so no model is ever a hard dependency for the system continuing to work.
That is not a house style. It is the only way we know to put a model into something that moves money and still sleep.
Yours could be
the seventh.
Capacity: 2–3 engagements per year
Start a conversation →Budget range, timeline, and whether IP or equity is negotiable — asked first, because they are the answers that decide.