2024 MIT Houston Symposium: Startup Exchange - Leela AI

CYRUS SHAOUL: Hi, everybody. I'm back. Great to be back here. And a lot of what I'm going to say, I think, parallels what Andy just said, which is that we've got to get better at understanding what's going on, and we need better data.

And so we've come up with a way of doing that for a uninstrumented parts of an operation. So yes, you can right now, do a lot of stuff if you have great data analysis with machines, but what about people? And so that's what we approached in our when we started this company. We are a Google Cloud partner, but we also work with other clouds. But Google, we've been working them first, and so they are also promoting our system to all their customers as well. So let's go to the next slide here, if I can-- OK, there we go.

So the big problem that we noticed when we started out talking to manufacturers is that a lot of parts of an operation are very much people centric. There are machines that they use, but people are using the machines. The fully automated parts are actually pretty much minority. And so you can get this kind of data like, oh, the lathe stopped. And so we have this overall equipment effectiveness of 34%.

This machine is running. That's great. But then there's all these other parts of the operation that are just not really giving good data. If they do, they may have people scanning things with a barcode scanner saying, I'm just going to start a build. And they scan something, and then an hour later, you get another scan saying, I just finished a build. Now, sometimes they forget to scan out, and then you have bad data.

So this data quality, when you're relying on people whose job it is, not really to enter data to be in charge of that, data quality is not so great. So had a problem where they got a reason code from a manual data entry that was not very helpful. Who's ever had an unhelpful reason code? Anyone? Yeah, OK. Omit that.

So what we built was a way to use off-the-shelf cameras to have our machine vision system recognize people and tools and parts and equipment and all sorts of other parts of the process and track them all the time. And the hard part was when we started out, the kind of technology that was available, you could track things, but it couldn't tell us what the relationships were.

So being MIT people, we said, I guess we're going to have to invent our own engine to understand relationships. And so we went to the drawing board, and for two years, we had to invent a new kind of AI system that could actually understand all the relationships, all the time between all these things. It's not easy, especially if you want to do it in real time.

And we had to also do it in real time because our customers want to know now. They don't want to know next week what happened today. They want to know now what's going on right now, because our customers are plant managers, line supervisors, operational leadership, continuous improvement leaders, quality leaders, safety leaders. All these people are our stakeholders, and so we had to have something that was fast.

So we used Google Cloud for the cloud operations right now. We can use other clouds too. But we also have an on-prem solution. And this has been very attractive to some of our customers because they have a place where there are a lot of people working that doesn't have a gigabit connection. Who has a place without a gigabit connection where you got a lot of people working? Anybody out there? Yeah.

And so without a gigabit connection, sending 100 streams of video to the cloud is not possible. But we can send it to a server here that processes the video locally and does a lot of the insight generation right there on a server on the local area network, and then can also communicate with the cloud.

And so again, having many ways to plug into this has been valuable. We had to build our own dashboard because a lot of our customers aren't using Tableau or Microsoft Power BI. They can't really build dashboards themselves. So we built a very nice dashboard that a lot of our customers are very happy with. If they don't like that, they can use APIs to pull our data and visualize it any way they want.

We also have our own alerting system, so we send out text messages and emails and other kinds of alert systems when our agents detect something that needs really rapid response. So that's the other thing that there's a lot of-- who's heard the word agent lately in this AI stuff? A lot of people talking about agents.

And so what we think of these are as visual agents, and that's sort of I think an up and coming frontier. You want an agent that not just can read through all the text and all the emails and pull out something for you, you want an agent that can be your eyes, so you can see when there's a safety issue coming up. Or you can see when there's these places where people should be working and no one is there. Maybe they got told to go somewhere else, but they shouldn't be doing something else. They should be back here.

When that stuff happens, our system can tell the right people, hey, check this out. There's something going on here that you need to look at. And that's been the moment for our customers. Like, wow, this agent is actually helping me out a lot. I really appreciate that. So that's what we offer to our customers.

And we've seen really clear ROI from doing this. These are the very clear ones that are quantifiable. We have a customer where they actually measured their capacity month to month after six months because they were able to identify a lot of the unexpected downtime.

This is manual downtime. This is time when people are just waiting around for parts. People are waiting around for a part to be moved into their area where they can start working on it because the previous team finished what they were doing. That kind of stuff added up, and that is continuing to increase their productivity.

So this is without buying any new capital equipment, no new machines, not hiring any new people. This is just everything as it was, but pulling out the wasted time using our agents.

We have customer also, some customers use both, so this one, different customer but they also do productivity stuff. They also do safety stuff. And this is a big cultural shift. We talked about cultural changes. Cultures of safety are important. And so to get that kind of change in culture, if you have a guardian angel, which is an AI watching for near-misses and accidents that are about to happen and then everybody knows that these near-misses have been adding up this week, they start changing the way they behave.

Because now, it's not just, I'm just doing whatever I want. No one's seeing what I'm doing. Now, I could die, and this thing is actually making me be better. So that's important.

This is the kind of data view that we have in our dashboard. You can see if anyone has ever analyzed like a performance of a machine, I'm not treating people like machines. I don't want to do that. But on the other hand, there's some really great science of manufacturing and science of operations that comes from thinking about things like, what is this time loss. What is the Pareto on the breakdown of time loss? We can do that now for a fully manual location.

So if you're building big things, big expensive things, and one day of delay is a giant problem for your manufacturing operations team, getting this kind of report daily about what our time loss was every day, it's a big deal. And so that's how we've been able to do all this kind of stuff, and you can see cycles and all that kind of stuff here. I'll do more demos of my booth over there, so come by afterwards, if you'd like to see more about what this does.

We are highly secure, and so that's gotten us into a lot of places where they're worried about data security, especially on prem and the cloud. That helps with security.

And why Houston? I think Texas is a great state because it's got this incredible ranking compared to California. It's the second one here in terms of patent filings and also, in terms of the metros, we got Dallas right there. So we've got a lot of great Texas power in high tech and AI, and I think this is a great place for Leda.ai to get going. We don't have any customers yet in Texas. We'd love to have some.

So thanks for your time today. There's my email. You can always email me, but please come and check us out, especially if you're involved in operations efficiency. High-mix is great. We love people who make one-of-a kind things, big things. That's who most of our customers are in high mix, low volume. So if that's what you do, please come on over. Happy to talk to you. Thank you very much.