娇色导航

Hi, everybody, welcome to DEMO, the show where we have companies come in that show us their latest products and features. Today, I'm joined by Mike Mason, he is the chief AI officer at Thoughtworks, frequent guest on our other shows. So welcome to the DEMO show. Thanks, Keith.

Good to be here. And you're gonna show us the Haven team assistant, right the and tell me what is Haven and what does it do.

So basically, there's a ton of hype right now in generative AI around what you can do with it when building software, right, especially code generation. So these AI's are spitting out lots of code for developers.

But we actually think that generative AI is useful across the software lifecycle not just for coding, for all the other things that the software team does when they're building, building software. And Haven is a team assistant to help you across all of those different roles and tasks.

And I'm going to show some of those today. Who is the main kind of role within a company that would that would benefit most from using Haven? Is it across the enterprise? Is it just developers? Like who can this help?

So it's software teams, so it's everybody who works on, you know, building software to create some business goal.

So, you know, that could be from people who are doing the requirements for the software, people who are designing user interfaces, people who are actually doing the coding, testing deployment, operations in production. So like the the full software lifecycle from kind of idea to production? Yeah. Okay.

And so let's jump in, then there's several features this has, I think you showed me a slide where it had bullet list after bullet lists every bullet list, but we're gonna focus on three cool features from from this. So let's go right into the demo. Sure thing. Okay.

So the first thing to talk about is the fact that Haven is really about helping teams do something we call knowledge engineering. And so what they're doing is figuring out how to express to an AI, what is the software that we're building?

Why are we building it, what is the technical architecture, all of that kind of stuff. And so there's a feature in here called Team knowledge.

And one thing I'll say about this is that there's a lot of demo where here, the intention is for people to take this software and customize it for their own teams.

So you're gonna see stuff running against GPT 3.5, GPT 4, this thing also works with Gemini on on Google and also on the Claude on the AWS. Is that really important that a company has access to multiple MLMs?

They don't have to have access to multiple, but it helps that this can work with whichever one they feel comfortable with. Because, you know, with AI, there's all these intellectual property concerns, all that kind of stuff.

So if a company has already figured out, Hey, we're comfortable using this particular system, Haven can work with. So within here, there's this demo that we're going to be using today, which is about EV charging stations.

So electric vehicle charging is like the problem domain that we're going to be showing off. And in this, you can say, hey, you're part of a team building software for EV charging stations. And this is kind of the kinds of components. This is why we're building it.

These are the features of the software, all of that kind of stuff. So basically, you're you're teaching the AI, what it is that you're building, so that it can interact with you in intelligent ways. The other thing we've got in here is Team file based knowledge.

So OCPP is a kind of technical documentation for EV charging. And so this allows you to access that as well. Okay. So the first thing we're going to do here is requirements breakdown.

So this is basically taking a high level idea of some functionality that you want to build in your software, and then breaking that down into requirements that a developer could actually go and implement. And we're going to use an AI to help us with that.

So the story we're working on here is implementing a secure and versatile payment processing system for EV charging.

And what the AI is doing here is it's taking that description, plus all of the other knowledge all of that context that we've put in, and he's coming up with ideas for features in the software that will be useful for, for this particular story.

And so it's it's starting to generate these things, Hey, maybe I should be able to register as a user, I can use different payment methods, all of that kind of stuff. Maybe you want to see real time charging status. That's a reasonable idea.

What's happening here, right is the AI is helping someone think about that requirements. So not just something that's obvious, but like coming at it from various different directions and getting that kind of AI creativity to help you out, okay.

And then what you can do is you can you can actually explore these things in a bit more detail. So you can say, Please give me more details.

And the AI will actually like go to town and flesh that out into something that's getting a bit bit more complete and getting something closer to what you can implement.

It's also asking important questions like what kinds of payment mechanisms do you want to take in this in this particular case? You can even then I'd say, just please make your own assumptions. You don't usually usually want to see that with it with AI.

Well, I mean, I think you, it's interesting, right? Like, you probably wouldn't just do this if you were doing this for real. But for the purposes of the demo, we can get the AI to fill in all the blanks for us, and come up with that that functionality.

All right. That's cool. And then this the second one, you know, I noticed you have a diagram here. I think this is a cool feature of the diagram analysis or diagram summary. Yeah.

So basically, we've got a diagram here, this is actually from AWS, just a random diagram of how an EV charging station might work. And these are all kind of the technical components to do with that.

And if you handed that to me, I'd be like, I have no idea what that is. So you're not technical, no person, so you wouldn't know it wouldn't know what that is so. So we can do something called technical diagram, description.

And we can upload an image, right, which is that same one that I was showing you just now. And we can describe that and we can say, you know, this is an AWS architecture for EV charging.

And what it's doing right now is it's going through the this is using GPT for vision. And it's kind of looking at that image, trying to analyze it understand what's actually in that. And once it's done that, we can then get the system to explain it to us.

If we're non technical, and we don't, we don't know exactly what that what that thing is, I'm gonna go off script just a little bit. But if you uploaded a picture of a cat, would it just be able to describe?

It would probably say tell me about the, you know, technical architecture of the cat? I'm not sure it would give you anything particularly useful. Alright, there we go. Okay, so did that. So it's basically described the whole, you know, as it's talking about the diagram.

And then what I'm doing here is I'm saying, hey, explain this diagram to me. And it says, Sure, great. And it tells me about the EV charging point, a load balancer, auto scaling all of these AWS components that were on the on the diagram there.

And basically, that's, you know, you can then drill down and ask for more detail. You can also do things like saying, hey, I want to implement this particular feature that the payment processing.

And you could say to the API, which architecture components are relevant, right for that, and it will help you understand that a little bit better?

Yeah, as we're going through some of these features, what would a user or a member of the software team do before a tool like this? What would they have to do? Would they just be doing a lot of other research?

Would they have to rely on other teammates to explain things? Yeah, that's a great question. They would generally have to rely on like the most expert person on the team to understand what are all the things on this diagram?

And what do they mean, the nice thing about building some of that knowledge into the AI is that you can then have some of that benefit of conversing with an expert, without requiring your team leads to spend all that time with you.

And you can use this for all the different roles on the team. So somebody who's an analyst, like trying to figure out what features are useful? They can they can do that. They can benefit from that exact same kind of expert AI. Right. Okay.

And then you can also then, after you've gotten the answer, you can follow up with questions, but I want to get to the next the next one. I believe it's called Knowledge chat?

Yeah, so basically, so I was talking about the big document with all of the technical specs for this particular protocol. So there's a bunch of different OCPP Yeah, exactly right. But you can build other stuff in here.

So for example, we've got you know, you can look at the Thoughtworks guide to agile software delivery, and use that as your knowledge base, and then ask questions about that if you wanted to. We've also got Martin Fowler's website in here. He's our chief scientist.

But if I look at the OCPP protocol, right, and I'm just going to bring that up for you, briefly here. So this is a 580 page, super dense, technical document, right? No way anybody's ever gonna read 580 pages worth of stuff.

It's got like architecture, diagrams, all sorts of stuff in that, right. So the point of this is to be able to ask a question. So here, if I'm working on that payment processing story, and I've got this big architecture document, I want to understand, is that actually useful.

So given that story about payment processing, is this OCPP thing going to help me at all. And so it's saying, not really got any specific information. But there is something about credit cards and debit cards in this dark.

And if I scroll down a bit, and I have a look at some of these things, you can see it's got page numbers, it's going to drill me into page 130.

And if I zoom in a little bit on this, you can see that this is an example of something in this technical document that has to do with payment processing.

So it would scan the entire document and then based on the question that you ask, it can tell you exactly where it can tell you is that just a keyword search is not keyword search? No, it's actually generative search.

So it's doing a useful kind of, you know, genAI style access to this knowledge. You know, we've got several different knowledge bases loaded into here.

One of them's for a medical record system, which has like, you know, 2000 pages worth of their their operational documentation for how to run the thing. And you can query that questions as well. Yeah.

So we got one more feature that you want to show this is about threat modeling or brainstorming. Yeah. So this is sort of modeling. And I'll kick it off. So we can we can talk about it while it's running.

But basically, I'm talking about who's using a feature, what are the assets that we're trying to protect, and roughly what's the data flow going on here. And Threat Modeling is important, because you don't want to do security just at the end of a project.

Ideally, you would like everybody on your team to be thinking about security, especially developers, but it's kind of a specialized activity. And so what we can do here is we can get the AI to help a developer, think about threat modeling for the feature that they're implementing. Okay.

And this is under the covers, this is using a particular style of threat modeling, called stride where you think about particular types of threats to the system.

And so here, the AI is coming up with things like spoofed identity with a stolen credit card, tampering with payment, data, all of that kind of stuff. So these are potential ways that there could be some kind of a threat to the system.

And this is specifically around the EV charging exactly, because we've given it all of that context about what the team is building, also the specific feature that we're working on, and then the AI is helping us look at that specific threat.

And then you can go into explore these things. And you can say, could you give me more details and mitigation? So this will actually sort of explain what is this threat? And how might you create ways to work around it? That's really cool.

Okay, and so, is there a pricing model? Is this subscription based? Okay, so or is this Do they implement certain parts of the platform. So this is at the moment, this is an accelerator for ThoughtWorks clients.

So if you're a client, we can bring it to you and you can use it, we're going to be releasing an open source version of it. And we're actually going to be splitting up the software from the knowledge pack.

So the knowledge pack is all of that embodiment of, you know, how you, you want the AI to prompt you.

And we are actually going to be embodying kind of 30 years worth of ThoughtWorks expertise in how to build enterprise software into a knowledge pack that's going to be proprietary, that won't be free. But the main software is going to be free.

And just since you mentioned it, can you explain to someone who might not know what Thoughtworks is? Sure, yeah. Happy to you know, you don't just create software for the heck of it.

No, we don't just create, it's like no, and I like to say we are the biggest tech company nobody's ever heard of, because we're 10,000 people globally, we build a ton of software for our clients, you know, airlines, ecommerce startups, or, you know, the full gamut of companies.

So for people that want more information about Haven, if they are Thoughtworks client, where can they go for more details, you can go to Thoughtworks.com. And we'll leave some links, probably in the video description. All right, Mike Mason. Thanks. Thanks again, and thanks for the demo.

Happy to be here. Thanks.