Introduction

I feel really inspired, so I'm gonna do a little quick bonus. Who wants to see something interesting that's a little techy too? Because I was gonna stick with the non-techy, but I've got something techy I can show you that kind of builds on what... What Miles just did.

Before I go there, so this is how you can reach me if you're interested in catching up with me afterwards.

The Fascination with AI Agents

I happen to be super interested in AI agents today. This year, in fact.

So I went ahead and self-published a book. It's all about AI agents in the enterprise.

And there's probably... Probably everyone in the valley right now and in other parts of the world are talking about this being the next form of software.

1So imagine autonomous software creating websites on the fly, doing transactions, interacting with customers, et cetera. So there's a lot of research you can do on your own.

But suffice to say, what we're talking about with agents is something new, something that you can basically give them instructions and they'll go off and do work on your behalf.

So it's a new area of research and applications and that's... That's how you can get ahold of me if you're interested in talking to me about agents.

Tech Demo: A Sneak Peek into Crew AI

Now, I will take you through a no-code environment in just a minute, but given that I just saw some code and I felt a little inspired while I was sitting there, I'm gonna show you a little bit of a bonus, and that is I'm not gonna expect that you'll be able to see this, but I'll make this a little bigger. I'll make this a little bit bigger and hopefully you'll be able to see that a little better. Okay, how's that?

So this, for those of you that are technically inclined and those of you that are business inclined as well, I'm going to just step into some techie stuff, which is offered up by an open source project called Crew AI. So Crew AI is getting a lot of attention. This happens to be one of those open source frameworks. evolving it continuously.

There are multiple frameworks out there. And they're kind of...

So basically, I will not be going into detail with you here today, but basically what you do is you create a role for your agent, you tell your agent what it's going to need to do, you give it a goal, and what effectively they're allowed to do.

So in this case, I have an example that we have a sales rep agent, we've got some backstory, I'm sorry. We also have a lead sales representative. This is a demo that you could probably find around as well.

And effectively, you... add some tasks. So you're gonna say, conduct an in-depth analysis of whoever you're submitting in company X and what's the expected output. So this is again, a very technology oriented way. that you can code if you happen to be interested at the level of code solutions.

In this case, there's a crew that gets kicked off, and that crew effectively has agents and it has tasks. If I come in here and I just run that, It uses a framework called React, response and act. And this is a new form of these agents kind of doing their work.

And again, you will have a little bit of trouble with the lighting in here to look at that, but basically it says, when it runs this, it says, here's my task, here's what I'm gonna do, think about it, then take the action. And I gave it tools to be able to search the internet, and I gave it a tool to do some analysis.

Okay, so it says I need to gather information, it goes off, it gathers a bunch of information, and behind the scenes, it's, transferring some of this information in memory from one agent to another. So, you know, there's a handoff of information that's happening and the other one goes off and completes an email.

So this obviously is done in a terminal setting and doesn't have a very good UI, but now this is at this level of kind of innovation that's happening, sort of builds off of what Myles was just showing and I could use a llama in the background as one of the models if I wanted to. In this case, I was using an open AI model. Okay, so that's a bonus.

Who finds crew AI interesting, by the way? Yeah, yeah, a bunch of you. I do too. I think that's a space to watch for sure.

Exploring No-code Platforms

Let's get into something that I'd like to show you now, a business level framework, business level solution that you could use.

MindStone Meetup Insights

And there's a lot of minds going around today. We're at the MindStone meetup.

There's a platform out in the West Coast called MindStudio. 1So MindStudio allows us to create workflows that access their access models. and you're also able to integrate very easily with other services such as Airtables or Slack or HubSpot, Zapier, all of that.

A quick show of hands, who's familiar with Zapier interfaces, APIs? Okay, good number of you. Let's take a look at what that looks like.

Let's take a look at what that looks like. Bear with me while I type. up here.

Navigating MindStudio's User-Friendly Interface

Okay, so again, I'm not affiliated with Mind Studio other than I'm actually using Mind Studio and the platform to build out some solutions that are becoming actual applications. Okay, so it's a very interesting, kind of easy to use business-like platform.

And effectively, you create process steps, and you decide what you want to have happen in every one of those process steps. And you can choose the underlying model that you'd like to use for each of these process steps. So it's pretty user-friendly, and you'd be surprised at what you can do.

I'll take a quick peek at their models page. So basically, There's a set of models that you can access. So all of these, as soon as a new model comes out, they pretty much make it available on their platform.

And we're gonna do something pretty quick. You can see that they've got their starter plans as everyone does, and they've got their enterprise plans coming. So let's do something here.

that I'm going to create. I'm going to create a new AI and it's going to be a blank one from scratch.

Okay, so this is their UI and you can change up the prompt, this is the system prompt. You can tell it what kind of assistant it is. And right now it just basically comes out of the box with the role of the assistant to help the user responding to the request, right?

Okay, so what do we have over here? What we have over here is now a start and an end. And I can come in here and I can say, I want some user input I'll try to do this on the fly here.

Short text. You can take in all kinds of input, but basically, it has a memory feature so that you can take some input, have input set as a variable, then you can use that content in other workflow steps. Really quite ingenious.

We'll just say company. What company will we research? This is obviously simplified in terms of what we might typically do, but if I come back here and I look at that, now I've got a company field that's gonna be populated.

Okay, and let's add another. Let's add another step in here and say, research the latest news about this company. There's different ways of doing that, but I'm going to use the variable that we previously received. And we're going to save that to a variable. All right. So we're going to save that. We're just going to call it result for the time being. All right. And we'll add one more. We'll add one more.

And now, obviously, this can be formatted in many different ways. And the system's going to send the output over here.

You say, how is this even possible that I'm gonna have a model that goes out and possibly searches the internet, right? Not exactly easy, except that they've recently added, perplexities models into their platform.

So now I can choose a perplexity model. Why don't I choose the large online one for argument's sake. I can decide how much I want to come through here and I'm going to preview the draft.

If all goes well, I should be. All right. So now I'm previewing it.

I'm previewing this application and it says, you know, let's see if the demo works for us here live through the local Wi-Fi. So over here, it's telling us what it's doing, how long it's taking, basically the backend. It's also saying how many tokens did you use? What did it cost you in terms of the usage, right?

And there it was. So I gave it no formatting in this case. I just let it format however it wanted. But as easy as that, I could have something that is now, you know, in a couple of steps doing some research, right?

Now I could also send it somewhere else. I can summarize it. I can connect it up to Zapier and send an email through my Gmail. All of these things are possible.

I can also embed this in another website.

No-code Versus Code-Oriented Solutions

So that's something that I won't necessarily show in demo mode, but when we take a look at this, we do have the ability to have this set up as an API that I can tap into, and I can also embed this in another website so I can embed this workflow somewhere else. Or I can call it, obviously, as I just said, through an API. So it's really pretty powerful and simple to set up, which is pretty magical.

I wanted to bring that as a contrast to how kind of tech-oriented you have to be and understanding in order to run something like CRU AI today. And there are UIs coming for that. But this is what's referred to as a no-code platform. And that no-code means you're effectively telling it natural language, what you want to have happen, and setting up the workflows.

And in this case, I have the automation here. Now, what's even interesting, and this came out recently, is that you can compare multiple models to see which one of these gives you a better result. So you can say, I want to use one from OpenAI, I want to use one from, I want to use Lama3, I want to use another one. And so you can pick and choose and run the same prompt against multiple different models and you can see how they behave differently.

Now, one more thing I'd like to show here is the following. All right. Let me take this. I'm just going to add like a little, call it a little menu item. Okay. And we're going to do review docs. We're just going to say review docs. I will take... In fact, I'm just going to... I'm going to do that. I'm going to leave that open. I'm going to say that will come over here. So now I've added a menu. This one will still be... Oops. Typing up here. I'll still leave that one as search.

Okay, so you've heard, who has heard of retrieval augmented generation, RAG? Okay, half of you.

It's a really important term in the AI world these days where you take a document and you can basically interact with a document. Examples are searching PDFs and interacting with them, like chat with your PDF sort of thing.

So in this platform, and I find it very powerful, is that I can add data sources. And let's just say I have some sales docs. And they can be internal documents if I'm setting this up for my organization.

And I'm going to upload, let's see if I can do these guys here. All right, so what it is basically doing here is it's uploading the file. It has an internal vector embeddings process, and it creates that in the background for you without you having to worry about any of these concerns about creating your own vector embeddings.

So again, very easy to use. And if I take a look at the preview here of one of the documents, okay, so this is a guide to sales prospecting. And what did it take out? So it basically extracted

extracted all the data on the fly. It created its chunks. For those of you that are in the weeds creating these vector embeddings, it created the chunks and they're your vectors.

So effectively, you've got an automated way of being able to interact with a document. Well, how does that actually happen? Here in their reg option, I'm gonna choose that particular data source

And I'm going to give it a couple of iterations here. I'm also going to preview this. Now let's do a new thread.

So our application has changed. It's got a little menu. Obviously, you can add a lot of UI sugar to this.

So let's do a review docs. I'm going to move the thread over here. And Tell me about sales funnels, something like that.

Now you can chat with documents and the end users have something easy to access and it comes up and it tells you where it's finding the information because it's got access to all the vector embeddings.

So in just a little while here, we've got a couple of workflows, we're using documents that users understand, and we have ways to chat with these documents. This is really interesting because it makes it very accessible for non-techies, business people to create their own workflows for their own applications. And as I said earlier, I'm using this for some more advanced workflows. You can have multiple different paths.

For those of you that are interested, if I come in here and I'll edit the AI. Let me just pick one of these guys. No, I'm just going to, just out of... run a function. We can browse some community functions here.

We have all kinds of potential integration from Airtable to run HTTP requests from other places. uh as well as uh you know your typical you know slack web hooks and and whatnot so so in effect you've got a no code environment but you can also expand it with some code if you want to and you can tap that tap into like web hooks from other sas providers so It's really interesting to me how quickly we've moved on from you had to roll your own rag necessarily to now you've got the ability to give non-techie users the potential to create these. And while you can certainly expand them and you can certainly, as I said, I've got some complicated ones that interact with... Zapier, Slack, Google Docs, a lot of data potentially moving around.

but it's very simple to get started. So I don't think in today's world, you don't have to be super techie to get some value out of these tools right away. Again, as I said, I'm using them. I don't have a particular other commercial interest in the platform, but I definitely wanted to show the potential of how easy it is for you to spin up some of these workflows. So hopefully that was interesting.

And you like what you see because there's kind of two elements, right? We brought something super technical, something, you know, even on the agent side, the crew AI stuff is pretty technical right now, but it's powerful. And then this is now an entryway for non-tech kind of business-oriented users.

Conclusion and Q&A

So with that, maybe I'll open the floor for a couple of questions before we go on to our next presenter.

Yes, sir. At the back. Thanks.