So in your experience, what are the common areas where software medical device companies tend to get their non-conformities or areas where maybe they need to go back and revisit? What have you seen?

How long have you got? Yeah, I think...

It probably remiss of me maybe not to start with.

I think clinical evaluation is a common area where we're raising findings.

And I think it kind of makes sense because I think generating high quality clinical data and forming a robust clinical evidence base for a device is definitely, I think, one of the more challenging aspects of bringing a device to market.

So I think...

Areas where we may commonly raise findings might be around the methodology of the activities that have been used to generate clinical data.

What makes the difference between a good PCCP rather than one that's just been made for the sake of it?

So mainly... we suppose like for example i work with some companies that are doing those devices like we have a camera that is making some pictures capturing some pictures and then identifying there is a cancer like for example when you do endoscopy or when you do whatever so this this kind of thing so which are more image related or other that are more signal related you you cough and then you identify what kind of cough is it or a stethoscope that measures the sound and then you are identifying if this is there is a disease etc What makes LLM different from those devices mainly? Is it the same or you are really having a different view when you receive a client and say, oh, I have a medical device with an LLM inside?

Yeah, I guess maybe what's interesting about those examples you provided is that the intended purpose, the scope of the intended purpose is quite specific.

It's analyzing a waveform from a stethoscope.

analyzing an image for like specific kind of like, uh, pathology.

And so I think one area where potentially like LLM based medical devices kind of deviate somewhat from, it kind of feels weird to say like traditional, um, but you know, like traditional kind of AI medical devices, I think is in this sort of like generative output.

So, um, The device isn't selecting from a fixed set of answers, right? It's constructing a response, potentially.

And so that output can be super broad and is generally fairly unbounded.

Is hallucination something that you are also talking about when we have this discussion with customers?