Matt Price

Second is resilience.

These systems now are becoming...

frontline for organizations, you know, concierges into the organization.

They have to be available.

You know, so if open AI goes down, you can't have customer care going down.

You know, so for real, we've built an incredibly resilient failover architecture across multiple models that can, um, that can, um, flip.

So we, we never go down.

We can, we can work off, um, many different AIs and that's actually useful from the innovation perspective as well, because

If we see capabilities improving in some of the base models, then we can flip between them very easily and take advantage of that.

So we have this layer orchestration, which is a multi-LLM model where we can change between them.

Obviously, associated with that, there's a lot of tooling we do about quality assurance and benchmarking and things like that.

Typically, they're curious, but they trust us to make the decision, which is why they typically partner with us.

They

Most of our clients have never done this before.

We've done it hundreds of times.

And so they trust us to do it.

And then, yeah, I mean, a lot of it is, you know, when we make changes, you know, we have each model is then tried and trusted or tried and tested.

to make sure that, you know, it's quality assured.

And if we move between models, it's going to give the expected outcomes and answers.

Yeah, I mean, I should make clear as well is that that accuracy level is the point, the small decimal points are not hallucinations either.