Introduction

From Clinician to AI Entrepreneur

Hello everyone, my name is Amir.

Thank you, Josh, for having me and having me back so soon.

I was here in December to give a bit of a story about how we came to meet and that was through TikTok, funnily enough.

So if you know that story, then great.

I'm not going to repeat it here today, but I'm really excited to give you a live demo.

So, you will have noticed that DR means I'm a medical doctor by background, which is an

unlikely place for me to end up here today.

I'm not a technician, I don't know how to code, and I never have.

But a couple of years ago, working on the front line of the NHS, when AI was going crazy,

I started to think to myself, God, there are so many inefficiencies, obvious things in

in our healthcare system that can be changed

with very basic tools.

Why are these not being implemented?

The tech exists, people are building this stuff,

but why is it not actually here at my fingertips?

So I kind of did the brave thing

of trying to address it myself

and went on this wonderful AI journey, startup journey.

And so I learned a few things

and now I'm at the point where I want to show you.

Setting the Stage: The NHS Context

First of all, any clinicians in the room actually?

Anyone from a healthcare background?

Oh my god, that's great, okay, that makes me really happy because when I was coming

here a few years ago I was the only person it felt like from a healthcare background

and I just wanted to scream and shout about this stuff to everyone else who I was working

with and everyone thought I was that weird AI guy.

So I want to show you how anyone can take very basic tools and use them to do something

practical in everyday life or healthcare or any other sector that doesn't revolve around

being a technician.

So everything will be live so we'll see how this goes together

but before we kick into it I want to ask everyone a question.

How the NHS Is Organised

How many trusts

trust being a collection of hospitals or community services for the NHS?

How many

do you think there are in the UK?

We've got 400 higher lower any takers?

Oh okay

Okay, so there's about 2 ,000 hospitals in the entire country,

which is a good guess,

and there's about 200 trusts in the entire country.

The Data and Digital Maturity Challenge

Now, what percentage of those hospitals do you think still use paper notes?

Not 100, you'll be glad to know, but it's...

Okay, so I'll move on to the next slide,

and I'm going to put up a bunch of stats which should explain it.

I won't go through all of them.

But there's 200 trusts, like I mentioned.

Those are rough estimates, obviously it's very hard to pinpoint these numbers, but 600

million patients in 2023 to 2024 made contact with our healthcare service.

Imagine that as a single data point, and imagine how many data points exist for each patient.

You start to grasp the idea of how much data there is.

Electronic patient records, which is where we store all of our information, this was,

you know, when I was in medical school less than 10 years ago, we used paper notes which

which was absurd, and now we're using paper notes in 70 % of hospitals, sorry, electronic

notes in 70 % of hospitals, but still, some things are paper only, and only about 25 %

of them are digitally mature electronic records.

The NHS app has been blowing up, 36 million people by 2025 are now on there, and it's

only going to evolve more, I hope, and the other place that data exists is in operational

use, so bed managers, theatre coordinators, anything that involves patients flowing between

one service to the next.

But the problem is, this data is a mess, frankly.

If you live

in East London and then you go to a West London hospital, your data doesn't carry over.

They

will know nothing about you unless you've been to that hospital before or if you come

holding your paper notes or whatever that's an issue obviously and even in one site the systems

that we use don't communicate with one another so i can be on the electronic patient record but if i

want to go on to my drug formulary and understand what dose i should prescribe specifically to this

one patient who has kidney disease and needs alterations i have to do it all manually now

it's a lot of steps it doesn't sound like it's a lot but it takes a lot of time and it could be

much more efficient.

Time Lost to Documentation

And documentation, which we've spoken about, 40 % of a clinician's time

is spent just on documentation.

So next time you're in A &E and you're waiting four hours

to 12 hours for a doctor, they're probably sat there documenting half of the time.

Let

that sink in.

Policy Momentum: The NHS Long-Term Plan

The good news is the NHS recognises this as a major problem.

So they've put this in their

ten -year plan which came out last year and one of the big points is converting

analog to digital so NHS app is one of their big initiatives and adopting

health tech in the wider community is a part of that as well as other things

such as going from hospital care to community care and sickness to

prevention which trust me we can do more of even myself included now I'm gonna

Live Build: Choosing a Problem to Solve

give you a couple options and in real time we're gonna go start up mode and

try and build something together so we've spoken about fragmented patient

patient data, documentation overload, and task chaos, task management in hospitals.

So can I get a show of hands for option one?

Which one do we want to build?

Option one?

Option two?

And option three?

Decision: Task Management

Okay, okay, task management it is, okay, let's go for it.

Research First: Using Perplexity and Prompt Engineering

So let's head over to first of all perplexity.

and find out a little bit about this.

Obviously, in the real world,

we would want to do interviews with clinicians

and people using these systems,

but here we'll just try and find out as much as we can

using deep research from Plurpexity.

And here I've got the pro version.

So, okay, dictation doesn't work here for some reason.

I'll just have to type, sorry.

I want to build a clinical task manager

used by doctors in the NHS.

This will be primarily for ward and A &E use.

It will allow doctors to input patient data and track the jobs required to be completed by each doctor in the team in order to manage that patient's care.

and then what I did before today was simply use chat GPT to engineer some prompts so here I'm

just going to copy and paste the prompts that I would use for perplexity to do a bit of deep

research give me a clear definition of this problem from a clinician's perspective

who experiences this problem now obviously I'll have a bit of insight but anyone else

not from healthcare wouldn't and then give me some non -AI workarounds that people are currently using

existing solutions obviously we want to do a competitive market analysis does anything like

this exist obviously we know things like notion exist for day -to -day admin and task management

but does anything for health care specifically exist and so on so deep research will take a few

minutes it's not as quick as a google ai chat gpt search but you can see there it's obviously going

through all of its resources and the thing i love about perplexity compared to chat gpt

I find it much easier to follow its line of thoughts.

You can actually go back and

see where it's grabbed this information from and if you want you can ask specifically for references in, you know, medical research specifically

that's really helpful.

As a clinician we're trained for everything to be evidence -based, so it's really hard for us to trust

things that just seem like they've appeared out of thin air.

Vision of an Ideal Clinical Flow

While that cooks, I

I want to run you through the ideal scenario now imagine this as a clinician patient comes to you

in A &E you have information about them that exists they scan it they give you their phone

you scan a QR code or whatever it may be and you know their entire past medical history their

allergies what medication they're on you don't need to ask all these questions for the first time

time.

You have a conversation human to human, you have that empathetic approach, you have that

real connection.

You give them the time that they deserve without worrying about scribbling down notes or rushing

back and remembering it all and typing it out.

1You get transcripts that automatically generate into clinical records.

With that, you get real -time decision support.

You get the notes that have specific recommendations for treatments based on

NICE guidelines, BNF, anything else, local guidelines, and that helps you make the

clinical decision, obviously with the clinician staying in the loop, using

their clinical judgment and their training to make the decision that's

best for the patient.

After all of that, the patient is handed over, whether it's

between services, so from the front door to the specialist, or whether they're being discharged

home to the community or wherever, a different hospital entirely.

Everything can exist in a streamlined manner, and this is the future that we're all hoping

for.

Translating Research into an MVP

So let's see what Perplexity has done for us.

It's still taking its time, so what I'm going to do in the meantime is go over and simply

use a chat GPT prompt to create the same thing on lovable because we're gonna

make it anyway aren't we we're not going to wait for oh there we go so it's given

us the problems briefly what we've described this is saying clinicians

spend 73 % of their time on admin which is even worse when you consider

referrals and prescriptions and everything taken into consideration who

suffer who suffers the most obviously junior doctors resident doctors who are

on the front line doing this stuff, this takes up most of the time, etc.

And then we can look at the gaps in existing solutions and we'll come on to these at the

end of the talk as well.

Prototyping with Lovable.dev

So using this information, we are going to go to lovable .dev to create an MVP which will

will be front -end only for the time being,

and then we will use that to pitch to investors

and build a full product later down the line.

I was hoping to dictate,

so I don't know why it's not working right now,

but anyway, I do this before going to lovable

because lovable, first of all,

Well, who's used lovable before?

Good.

Okay.

So a lot of people have.

In my experience, if you go there, you just tell it exactly what you want in plain language,

it works well, but there's a huge difference when you speak to it in specifics.

And if you tell it in layman's tech terminology, specifically what you want it to do in the

front end and the back end, and the best way that I've been able to do that is simply use

use an AI to generate that language for me because I don't speak it from my background.

So again, this might take a few minutes.

So I'm going to ask you another question in the meantime.

Say we build this product, who would feel comfortable if I start using it in a hospital

tomorrow?

Okay.

That's not the answer I was expecting.

Why?

Why?

So obviously there's risks of the tools hallucinating, there's the biases that we

might be unaware of, and it's not clinically validated, we don't know if it's actually

safe.

Take another product, for example, somewhere that AI is making a lot of progress in healthcare

is with radiology, x -rays, CT scans.

They are training models not just based on existing

images and diagnoses, but also using tracking software of the radiologist's eyes to see how

the clinician scans the images.

The AI is going to be able to do a radiologist's job to a great

extent, but still not being adopted because do we trust it?

What if it comes up with a

hallucination or a misfire?

Let's see what perplexity has given us.

I'm not even going

going to read this, I'm just going to copy and paste it straight into lovable and see

what it comes out with.

Now again, this will take a bit of time, but you can see what's happening in the background.

It's already put out all of its components and it's starting to code.

I can't read this.

If anyone can, good for you.

The other thing that they've started to do is integrate a lot of different platforms

into lovable.

So Shopify you can now use directly on lovable.

If you're doing something B2C, selling consumer products, you can integrate Shopify into it,

and it will take care of your entire marketplace.

They've got their own cloud software.

You can link up Superbase to it to manage your back end, your databases, and your authentication

of users.

No-Code to Startup: Integrations and Limits

So the startup that I built in medical education, I did it entirely on Lovable.

and it was only when we started building our own LLM

that I had to bring in external engineers

to help me with that

but it shows you how much can be done

with very basic tools

which today we're calling it basic

last year this would have been revolutionary

so here it's requiring us to connect it

to a super -based project of course

so this is where the complexities add

Plan B: The Patient Passport Demo

Okay, so instead of doing all that, I'm going to show you something else that I prepared

just before because I anticipated some friction.

Take one of the other examples.

This is a patient passport, okay?

I did the exact same thing to generate a prompt and plugged it into lovable and it gave me

this from the first prompt.

I didn't have to go back and edit it at all.

The patient can input all of their information right onto this passport.

They can have an emergency card so when they show up to an A &E we know everything critical

about them right off the bat.

They can put in their medicines right into here and then they can track their own history

and then they can come to us with that and we can scan it in a minute, hopefully upload

it onto our system and use it.

And this is what I'm talking about.

Imagine if you could generate a QR code that gives you access to all this information without

without having to repeat it every single time.

So something that I don't love about this is that the user interface when you come onto

it isn't very nice.

It's very clunky.

Imagine a patient is using this.

You want it to be a bit more obvious where to go and what to do.

Improving UX with Conversational Edits

So instead of going between the different tools, instead of going between perplexity

and back to Lovable, I'm just gonna chat to Lovable.

Hey, I want to create a more user -friendly

homepage that any patient, brackets,

from any cultural or education background

can easily interpret.

So now Lovable itself is going to suggest

suggest the changes that it will then implement itself.

So I'm talking to this as if I would talk to a CTO or anyone else.

It will just say, okay, let me look at the app first of all.

Let me see what's going on.

It's even taking a screenshot so it knows what the home page looks like.

And then it's going to start listing off why don't we do X, Y, Z.

And if I'm happy with it, all I have to do is click a button that says implement plan.

And the rest is taken care of.

So, I'll come back to that in a second.

Telling the Story: Building an Investor Deck

Now we want to take this idea to investors.

Create a pitch, a slide deck that will be presented to investors.

This should be six slides long.

I will copy and paste the output directly into gamma .app.

Gamma .app was mentioned earlier by Josh, and if you haven't started using it yet, pay attention,

please.

It's cut down so much of my time.

And the reason I know about it is from coming to this event in the past.

So remember what we spoke about here?

Unlovable it said let's do all of these changes, for example, accessibility improvements, bottom

navigation enhancements.

Cool.

I trust you.

plan approved let's just see what it does and then while that's working i'm going to tell you a bit

Clinical Safety, Trust, and Regulation

more about the clinical safety stuff who's seen this what do we think about this it's very

interesting isn't it now i'm up here talking about all these ai health tech tools and companies

what's to say that this isn't going to come and abolish everyone in the market tomorrow

There's a good argument for it because obviously they have the customer base already.

People have been using them for a couple of years and, you know, we would refer to Google

as Dr.

Google because patients would research their symptoms before coming in to see us,

whereas now they've got open AI, they've got anthropic.

And it's going to make things very interesting, I'm not going to go into too much depth because

I think it's too early to really say how these will unfold in the future, but right now these

these are not classified as medical devices in the UK,

which really makes you think, okay,

can we trust, but not only trust,

can we actually use this information in clinical practice?

I think it's gonna be a while before we get to that point,

but this is something to keep an eye on for now.

Generating the Deck in Gamma.app

Okay, let's take our output, which Gamma has given us,

and I'm not gonna copy and paste the whole thing,

I'm just going to copy and paste this text part, and it's literally spelled out slide

for slide what text our investor pitch should have.

So I'm just going to take that over to Gamma, generate new using AI, describe what you would

like to make, and I can select six cards just because that's what we said, and in real life

I would obviously skim through this, make sure all the information is correct, but once

Once again, we're doing this as a live demo under time constraints, so I'm just going

to copy and paste it blindly.

I can select specifically how much information I want to be present on each slide.

So for example, I'm just going to say concise because investors might not be bothered to

read too much.

I'm going to select this prism theme because everyone seems to be using that in the startup

world and it's shiny and fancy and nice, and we're just going to generate.

and I'd quite like this part of Gamma because it might be a bit of a gimmick

but it literally shows you all the text being typed out as it happens

and it just makes you think, oh wow, this is super futuristic and exciting

and I'm doing a lot of work just by typing in some prompts

and suddenly you've built a startup and a pitch deck in a matter of an hour.

Key Takeaways and Closing

So you can see how people are doing a lot of amazing stuff

in the health tech world with this kind of tools but I want to hammer home this

final message once again the barrier to adoption of this technology isn't the

technology or the lack of it it isn't the lack of skill or the lack of

awareness of AI or might be in healthcare to some extent but the real

The Real Barriers: Safety and Interoperability

challenge is the clinical safety and the interoperability and what I mean by that

that is, the tech that is built in the outside world, it's agile, you can make it quickly,

you have teams that are working around real life data points, you're testing it in real

time, you're iterating it and you're constantly working based on feedback.

In healthcare it's much more challenging because you have to build clinical safety compliance

into your work stream.

You need to build with evidence, you need to build with compliance standards, there's

this thing called the DTAC, the digital technology assessment criteria.

And you have to make sure

that everything is secure.

Because until you tick all of these boxes, no NHS body, no procurement

office is going to pay you a contract.

No one is going to sign anything with you until you have

all of these assurances.

Practical Advice and Support

So take away what I've learned.

I'm a clinical safety officer and health

tech consultant in my time now.

So feel free to connect with me if you have any questions.

I'm

A Hopeful Outlook

more than happy to help but what I want to really leave you all with is maybe a bit of hope a bit of

inspiration that our current healthcare system might be on its hands and knees at some points

but I think with this tech we can throw a lifeline and I believe it's not going to be long before we

see a healthcare service like this where we have futuristic holograms and real -life AI clones of

human beings that can predict our diseases and prevent them from happening in the future so

thank you so much and happy to take questions at the end

Break

We'll be right back.