Where it should live, where AI belongs, what it costs, and the costs nobody sees today.
It works as a demo. It runs on a hobby host in another country, with no backups you can wind back and no record of what the AI was told.
Prices are DigitalOcean's published rates in US dollars for their Sydney data centre.
Someone deletes the wrong patient at 3pm.
Balanced is the cheapest option where a mistake costs minutes, not months. Big is there when the clinic outgrows it.
Everything with one right answer becomes a rule. AI stays where judgement is needed.
A doctor still signs off every one of the four.
For two of the three reports, yes, the top tier is overkill. The doctor signs off every one, and software checks the numbers first. Dr Ashraf compares the first 20 reports blind, and the cheaper tier stays unless the difference is obvious.
When a report comes back wrong, per 100 reviews.
The AI bill for a year is about $35. Rework on 100 reviews costs forty times that in people's time. That is where the money is.
| Records | AI, to process them | Hosting, per month | Today's prototype |
|---|---|---|---|
| 1,000 | $35 | $67 | Works |
| 10,000 | $353 | $67 | Pages drag |
| 100,000 | $3,530 | $97 | Barely usable |
| 1,000,000 | $35,300 | $260 to $320 | Does not open |
AI is charged per job, over years. Hosting is charged per month. Neither is the big number.
A system that keeps patient data in Australia, can be wound back, spends AI where judgement is needed, and shows its working.