Explainers, and the receipts behind them
Life at the forefront of applied AI
I explain how AI actually works, in a domain where being wrong has consequences. Co-founder of Clinara, voice AI in production with aesthetics clinics. Before that, five years building Comet CX to twenty people. Every video here has a written twin.
Recent
All writing- Aug 2026Explainer + written twinHow evals actually work, and why yours are lying to youA pass rate is a summary statistic over a distribution you chose. Here is the mechanism behind the eval framework we run on Clinara's transcription pipeline, and the three places that distribution quietly stops matching a real consultation.Evals12 min
- Aug 2026EssayFour things our beta clinics taught us that we got wrongWe built for practitioners and the front desk became the power user. Transcription turned out to be the commodity and the private coaching was the reason clinics stayed. The full list, including the one we're still arguing about.Clinara9 min
- Jul 2026Explainer + written twinWhere the data lives, and why it decided our stackOnshore storage and inference in AWS Sydney was a hard requirement before it was an architecture. What that constraint ruled out, what it cost, and how the managed alternatives compared when we actually measured them.Regulated AI14 min
- Jun 2026EssayCustomisation isn't a settings pageEvery clinic has its own note format and none of them are being precious about it — it's how the business runs. Pushing that into the deepest layer of the product rather than the surface changed what we could sell.Product11 min
“Have the opinion before you script. If I can’t finish the sentence ‘and here’s what I think follows from that,’ the piece isn’t ready.”
Work with me
Three reasons people get in touch. All of them go to the same inbox, and I answer the specific ones first.
Advisory
Getting AI into production somewhere it matters
Discovery, architecture, evals, and the data-residency questions that decide the stack before anyone writes code. Twelve years of enterprise delivery behind it. A handful of engagements a year, usually a day a month.
Start a conversationHiring
Senior technical roles
I'm building Clinara and not looking, but I read everything. If the role involves AI in a domain where being wrong has consequences, say so in the first line and I'll reply properly.
Send the detailsJust say hello
Disagree with something I wrote
Corrections are the best mail I get. If a number in an explainer is wrong, or its source doesn't say what I claimed it says, tell me and I'll amend the page. Same goes for anyone building in AI in Australia.
hello@robscott.dev