AI-Powered Infrastructure:Building More with Fewer Servers & Nodes | KC Pele @Web3LagosCon

I'm Casey Pelle by name. Before we start, let's just play a little uh game here. So, last year and the previous year I've been advocating about uh AI. So, I don't know. I just want to find out which of you still write still convert your Figma design to code by hand.

If you are guilty of that, just signify. I think we'll be having a deliverance class after this session cuz why would you be doing that in this era? So all I'm just trying to say is that we have many servers few blockchain node and how to make it possible. If you notice all the speakers have been talking about technologies built on the blockchain. What if we now have those uh blockchain nodes here in Nigeria and not somewhere far or abstract?

What if we have a place in Lagos? I believe for me everything should not be in Lagos state. So that's just what I believe in. But then for now what if we have servers or these blockchain nodes that we are always talking about here in Lagos or in Africa but the good news is that we already have them. We have one in Lagos.

We have one in Niger state which is in Mina currently running. Those are blockchain infrastructures that we have. So I'll be walking us through how we actually made that possible and how AI can be of help to you. So if you have applications running you're building softwares you don't actually currently now we all know that developers your job is that at risk right right yeah so with this I'll just be showing if you're a founder a developer content creator all I'm just trying to say is use AI probably next year by the time we get here a gate fee or your registration slot thought would be if you're making use of any paid AI you get access into that but if you don't have an AI subscription because we have to start enforcing that it should be something that you use every day which is what AI so we have many servers few blockchain node how to make it possible with AI currently we've heard of AI agents right we all know about so how many of us know about um open claw which is claude bots, grock bots, hammies agent. How many of us know about that?

Good. So do you know that with this now you can have your own AI that does everything for you both your emails devops engineering work all you need to do is to automate the process like what hold said the thing is with his idea now he can just sit here chat with his telegram bots and then tell it to either add xyz future and then he sits here review the PR while we are here and then that gets shipped to production. Right? So those are some of the uses for AI. But then let's look at some of the bottlenecks and why I said many servers few blockchain nodes.

As we can see the bottleneck is operations and not the hardware. Currently with the new infrastructure we have for blockchain. You don't need um huge servers or complex system to run a blockchain node. With your laptops, your MacBook, your phones, you can actually use that as servers and also as blockchain node. So hardware is cheap, attention is not.

Every node you run multiplies the operational surface area and small team feels this first. So as we all know some of us are trying to cut down cost on our teams right. So with AI you can have that and then what it does is that it manages the state the pressure on your servers despite rates and some of those complexities. Recently uh last year we did a workshop on blockchain nodes where we educate people on how to run nodes. Now you don't actually you need that don't quote me wrong you need that but then with AI you can simply host these services and run the blockchain nodes you prefer and you also end from it I'm currently running uh near blockchain and all I do okay I can't show my telegram chat the only thing I do I just ask it the health status of the blockchain node it gets the log SSH into the servers.

I can be here in Mina and it's running in Niger state 24 hours light and then all I just ask is how is it performing? It just gets the log check the analysis if it's down by probably maybe system memory or just a crash. The AI would fix it itself and all I just do is just to confirm and I'm good to go and I'm also aing from that too. So those are some of the things AI actually helps us to do. So every new machine adds a load.

Each time we decide to add a new server, which is for now I'm running just one. If I decide to add the second one, it's a whole lot a different ball game entirely because I have to install set up and all I just need to do is to give the AI the SSH login and tell it to run near blockchain and I'm good to go. So those are some of the use cases we are currently making use of. So let me just take a quick break here now just to ask how many of us know about um clo design. Clo design do you use it?

Oh that's good. Amazing. How many of us know about um open design? Open design. I think I have some open source folks here.

So these are some of the tools. One thing I like to do is that when I give you the paid version, I like to give you the free versions too. So those are some of the tools you can run and it's very easy. All you need to do, you can go on railway render and any cloud service provider you have just they they now have um AI integrated into it. just tell it to install some of these open-source packages and then you get to use these tools for free without paying any of the services.

So now let's look at AI as the operation assistant. So it observes, understand, acts and then it learn is a continuous cycle. It's not just a a bot. It's beyond a bot. So first of all, it observes which is the first step.

You set up this um servers. Then it understand what you actually want to do because the AI has a broad knowledge of this technology. Sometimes you can just tell it to search the internet and then it would make use of the knowledge it has found and then it acts. What you just need to do is to tell it install XYZ do XYZ and then the AI gets it done for you and it's learn currently on my telegram chat I receive an email summary by 6 and by 5 and also my node status if anyone is down the AI by itself fix it so I don't need to tell it oh do this do that don't do this it just knows that this is supposed to be running and then it fix it. So now we have the reference architecture AI assisted development.

We have the act and observe the AI ops and oversight and the core infrastructure the runbooks deployment node control logs and the topology. Most of these things you're just seeing it's like oh it's too much for AI. It's never too much. You just let the AI have the logs. It gives you in plain English what that is meant to do and why probably your um systems are crashing.

And then we have the coordinators memory guardrails with human policies and accountability. This is just for some of the action cuz we might be scared. Oh, AI might delete everything on the server. It might crash. No, some for some actions you get a confirmation before you approve.

And then the core infrastructure we have the notes, the server, database and your network. So these are just some of the workflows that you would actually use when you have your um when you decide to run your AI infrastructure. So we monitor what is true and not what looks green. Sometimes some of our applications they look functional but then there are bugs in it. So but then with AI you get to see what is actually functioning.

Verify the real states alert on meaningful change and three-stage classification for alert on meaningful change. currently on my system. What I just do is that because I have public IPs that randomly change, I just set up the AI that each time it change notify me and also change that on Cloudflare. That enables me to continuously run these services 24 hours. So even if the IP goes down, comes on whatever happens, I'm constantly in the loop.

So now turn logs into diagnosis. This is something that um AI likes. So I won't really spend much time here. As soon as you tell the AI what you want, the AI gets it done for you. So assisted application and hosting.

So this is just um a side note what we do. We self-host platforms. You have your uh developer apps AI operation layer. You have Kify or Doy. These are open-source services like AWS, render, um, and other platforms.

Uh, Vasel, we are all common with Vasel, right? So, Kifify and deploy those are open-source services and all you need to do, you don't need to learn or read their whole documentation. Just tell the AI, deploy this application for me, and then you have your own vessel local to you. Then we have containers and public end points. Now proof of points recovering from out of memory crash.

So sometimes these are some of the setbacks we do have when using AI because on the same system you have your blockchains running and then you also have this AI agent running. Sometimes you do have conflict but then you can use your AI agent to detect, diagnose, approve, recover and also because we have these agents there are now being designed for them to remember what you have told them before. It is now easier for them to go over the steps just in case it happens again. So you don't need to start explaining yourself. Oh, remember what we did 2 weeks ago.

Remember what we did four weeks. No, the AI now has a memory of itself and it also remembers migration without losing context. So client migration and high-risk version change, config drift and sync gaps can silently degrade a node for days. These are just some of the overhauls that you might have. But I'm here to tell you that AI can easily facilitate such um development for you.

Support the apps around the nodes DNS proxy full dependency awareness trace the real root cause of such um um of such issues when it arises. So we now have the persistent memory is the on call what handbooks. We have our policy incidents procedure which is the proven run books and then we have the topology and the owners what runs where and who owns it. So sometimes you might think oh the AI is now in control of everything. No, you still own your infrastructure.

The AI is just there for you as a DevOps uh engineer. Please the DevOps engineers don't come after me. I'm not trying to replace you. I'm just saying AI is here to do some of what you're trying to do for us. Now our multi- aent archite architect orchestration sorry monitoring agent we have our coding agent which is our developers are here I also do this a lot I build application please if you are still telling your AI agent make this application very beautiful make it attractive don't write AI slop no we've gone past that there are hacks and tips that you can use to build application that does not look like every other um um AI generated software because now as soon as you tell an AI agent build me a front- end application there are three colors you would get is either blue and white light brown and dark and also orange with a touch of yellow in it.

Those are the colors you would get from AI. But then when you need something solid is beyond telling it make it beautiful, make it attractive, oh it should be impressive. Those words you would still get you a an AI slop. Yes, that's what we call it. So you would still get AI slop.

There are hearts to it. Then we have shared state verification with um agent. Then we have infrastructure agent. Then we have the central coordinators. So these are all shared state.

There's no opportunity for me to show a demo where I would have assign one agent to build an application for us and then it would assign it some another agent would deploy the application on um docloy. another agent would get us the URL and all we just need to do is just wait and then get our final application all hooked and set. So yeah, how many of us are still by coding? Okay. All right.

So I guess my time is up. All right. So we have report, recommend, approve action, reversible automation and also selfheal. So these are some of the things you get with AI. You don't need to stress in this era.

Any idea you have, I encourage you just build it, ship it to production and let the world know what you can build. Thank you.