Videos and Transcripts
Three short walkthroughs of the platform's AI layer in practice — a complete booking in conversation, the client-managed knowledge layer, and why lean organisations move quickly on platform frameworks. Full transcripts below, lightly edited for clarity; demo locations anonymised.
4 min · Watch the video →
This is the operating model behind the deterministic policy engine and the client-managed knowledge layer: the organisation owns its policy content and can change it immediately; the engine that decides bookings never guesses. The Policy Builder produces the pre-deconflicted document this video argues for.
For those for whom a picture paints a thousand words, this may turn the light bulbs on like it has for me. So here I am — I'm a client: an airline, corporate travel, whatever it might be. And what I wish to do is update my AI. Now I'm responsible, as the client, for the system prompt. I can change all of this. I'm also responsible for things like temperature and so on — these are starting to look at the guardrails. The vendor is responsible for choosing the model for the back end, Azure, all of that sort of thing. But as a client, I'm responsible for all of the business-rule layer that will be applied to the outcomes.
One of the ways of looking at this is the documents that you upload. If you have an updated policy document, you put it in here and out it pops on the far end. So now the organisation is responsible for uploading consistent documents. If you put contradictory documents in there, then your outcomes will be contradictory.
So how does this play out? Some people are deterministic — and yes, when I'm booking travel, I can choose to use the deterministic booking widget that I've always been familiar with, and that's fine. But now I'm going to come into the AI travel assistant, and I'm going to be quite specific about what I say here. I'm going to say: can you look up a policy regarding rubrics? Although I want to make a travel booking, I wish to understand some company policy before I do that. It goes off and it looks across — and it's coming up with these documents we just saw. If I want to drill down into these more deeply, I can do that; I can see exactly what's going on.
Instead I'm just going to say: thanks — I'd like to book from the offshore island to the mainland, tomorrow, just me, private. And it goes off and has a think about that. Happy with that — and book. And it's confirmed. But now I'm going to say: I meant to make a subsidised private travel booking. And it knows all about this too.
This is really changing the whole concept of that black box that was set up by the consultant ten years ago, where they were responsible for everything. Now the organisation that can manage the cultural change can create empowered staff who can respond to business requirement changes and customer changes immediately.
90 sec · Watch the video →
Note the key line: the assistant does not decide — the booking is validated against the corporate travel rules before anything is committed. The assistant is a client of the policy layer, exactly as described in the API and ontology reference.
What does a corporate travel AI booking look like according to UnityTrip? Well, this is where we've got to. I'm not going to use the widget — I'm going to come in here and say I would like to fly from the offshore island to the mainland, date of travel tomorrow. Yep, that's fine, that'll do me, thanks.
It's having a little bit of a think about that — because of course there are corporate travel rules in play — and it's made some estimates where I wasn't clear about what I wanted. Just me, thanks. So it's now coming back: booking validated, just one traveller. Shall I go and book now? Yep, that's fine, thanks. Done.
The thing is that all of this leverages off the back end: our system prompts, our conversations that you can go in and look at, held as metadata. There's a lot going on here.
90 sec · Watch the video →
This is the knowledge half of the doctrine — frontier AI at build time, determinism at run time — described on the architecture page: the client’s own documents ground what the assistant says, while the engine decides deterministically.
You're asking how UnityTrip can make so much progress so rapidly in the whole AI space — so here's a little bit of background. You can see we're in Microsoft Azure's AI services. The last thing we looked at was contextual uploaded documents for a client. I'm going to come into AI Search, click through into a subscription, go to the indexes, and pick a particular index.
I'll type in rubrics — like the example we showed — and here's what's in the background: it just has all these tools, making all of this available. We don't need to build this ourselves; we just need to take advantage of what's already there. Or maybe subsidised travel — here we go, here's that document.
So, as a little bit of a why: why are lean organisations able to make so much progress? Because they're able to sit on top of frameworks like Azure.
See It Yourself
The assistant converses; the deterministic engine decides. That split is the whole doctrine.