I am Rhys Morris. I am founder and CEO of TRM Flex. We are an event operations platform.

From Event Operations to TRM Flex

My background is I have run global events for 15 years and I had a career change. Back in October last year, I left my corporate role because my father was terminally ill and he passed away in January this year. But from that experience and the conversations I had, he said to me, go out and do something you love and you want to do. You've always wanted a product and you've always wanted to help people. And this is where TRM Flex came from. It's my 15 years of knowledge downloaded into this platform.

What I'm going to talk about tonight is how I built it and the different approaches I had to it.

Turning Experience into a Product

just a caveat i'm the ceo and founder i am not the cto and i don't pretend to be technical so i hope you will enjoy this tonight for me i already had all the information it was in my head i had all the documentation on my laptops 15 years of briefs all the different spreadsheets

so what i started doing was using voice chat function within ai and uploading my information to start mapping out what I wanted that platform to look like. Good old ChatGPT was my friend, and we started working.

Now, as we've here tonight, and as everyone knows, AIs hallucinate. They think they've given you the best idea in the world and solution to the product.

And then when you look through it, it will go, two more things, you can do this to make it really good. This was driving me nuts.

Using Multiple AI Tools to Build an MVP

So instead what I did and my approach was we have the first initial AI to start mapping it out, then I'd bring in a second, Claude, to check what the first AI had said.

Then I built on Lovable, yes I vibe coded it, Lovable is really fun, it's really cool, I got to a diamond level level 7 vibe converter, whatever that means. But Lovable had its restriction too.

But in essence, I just used these three and I'm sat in the middle, copying and pasting, understanding and trying to get to a point where I thought it was good enough to deploy it via Lovable.

Lovable was my builder so I could see it, so I could start getting that MVP out there.

The Limits of Vibe Coding

Now the one thing with Lovable is it does not have staging hands up how many of you have built with ai in here anyone use lovable no yes okay again i learned all this planning development staging production to those of you who are tech you'll understand this is second nature it's 101 it's like first day of school to me brand new world staging was non -existent now what does staging mean it's your rehearsal

space for those non -tech people in the room so the problem was I could have my idea I could take it through I can run it around the system it will get all the way to the end I then start going and speaking to friends family and potential clients and one change that I've done in the morning has resulted in me breaking something on the system in the afternoon. It was infuriating.

Why Staging Matters

By this point, I knew I was out of my depth and I needed to bring on my CTO. That happened within about three months, along with the CFO as well.

So the goal was to look at how I had been using BitBuildit and how I'd managed to get so far.

Building a Safer Development Process

My CTO's first email to me after he reviewed TRI Flex was he didn't know whether to be horrified or impressed because I had built so much but we needed to ensure that the platform could keep growing at the pace I wanted to but to be safe and that we had full control over it so what

you'll notice now I'm out to the middle there's no more copy and pasting and

Independent AI Review and Testing

this is our way that we build we have the idea the request that goes in the the AIs will start building that plan out and then we have built our AIs themselves to start checking, running tests, running all the checks on what we are looking to do and looping around.

The big thing is the AI who starts writing never marks its own work. We always get that second AI in, we get that third AI in as well. well.

Then finally once we are happy and we've said this is good enough and my CTO has gone through it, we will move it on to development where the idea will be developed out in code. From there we will move it on to staging where we will test it rigorously and then finally once we are happy we will put it through to production and that is what you see on screen.

screen but staging that is where we find a lot of the problems and all we will do is mark up what has gone wrong and we will send it back through this time and time again now one aspect on what

Reducing Complexity in Event Check-In

we were looking at when you guys came in tonight and this is now going into my knowledge field events when you guys came in tonight or any event you have to be checked in how you checked in probably QR code scanner tonight it might have been clicked on a computer you might have self -checked in what I wanted to do was map out how do we achieve that within the system for all the

different types of events we put it in on the plan we just saw the AI worked it through and came out with this amazing plan but what it had done is built four different routes into the system Why is that bad? Because we need to have one source of truth, one check -in, and that we can map who is actually on site.

One Source of Truth

What we do then, or what we started doing, was looking at why was this going wrong? Are we creating too complex of jobs for our AI system to deal with, and do we need to strip it back? so what we do now when we start writing our plan and what the next feature is or what we want to develop we put it into three these three areas what is the job is it essential for what you're

building is it optional or is it noise now taking that example of the check -in what we want to know is who is actually on site here tonight what is optional is how you get them in through the door that each one is valid and the noise has come from the AI itself in creating four different routes of mapping it into our system.

Essential, Optional, and Noise

From that learning we began writing out these rules based off the maps that were being created and hard coding it in so our AIs would stop drifting.

Creating Rules That Keep AI on Track

You can see three of our rules on the screen here that we have got within our system and we are developing these over and over again to ensure we are constantly learning we've also built out advisors as well

who can help us and we look not only from our own knowledge base but we will go out and speak to people who can help us develop that app now we ran that scenario back through once we'd inputted this update the ai ran it again on the maps you saw earlier it kicked it all the way through and down here these are the actual logs from that ticket as it's going through the system

eventually the AI itself got to the point where we could review it and we were happy with it and when we tested it we got the result we wanted one source of truth straight it it's complex

out there what I'm saying here is probably second nature to most of you I am in my AI apprenticeship shift on what we're doing and what we're building. If it's been helpful, fantastic.

Key Lessons for Building with AI

But there's three key takeaways that I hope you can take from it.

Explain It Like I’m Five

Who knows what ELI 5 stands for or who has used it? Put your hands up. Expected a lot more than that. Right.

My CTO told me this and I thought he was taking the piss in September. ELI 5 means explain it like I'm five.

It's been around since since I think 2011, it was used on Reddit, and basically it tells you what is just outputted as if you're five years old. Don't be offended by that. It just makes life so much simpler if you do not have that technical understanding. I use that daily.

Give it a go when you're next using an AI and it's outputted something that you just don't understand and you've asked it three or four times.

Strip It Back and Challenge the Output

the final two strip it back whatever your idea is whatever you're trying to build strip back that idea to those three steps I mentioned earlier is it essential is it optional or is it just noise could it be your own ego get it in the way and then finally argue it

never trust the AI on its initial output always test it always check it get other AIs to come back at you with it.

And finally, like I said, EI5 it. Give it a go tomorrow.

A New Chapter with AI

I'm really enjoying my journey with AI and it has allowed me to build a system that I'm really, really proud of.

It might be right, it might be wrong, but for me, it's my new life and I love it.

I'm Rhys Morris, founder and CEO of TRM Flex. Thank you.