Safer, Faster Public Transportation: AC Transit’s AI-Powered Upgrade with Hayden AI - Ep. 290

0:10Welcome to the NVIDIA AI Podcast. I'm Noah Kravitz. Join us at the world's premier AI conference. GTC San Jose is online and in person March 16th through the 19th. From physical AI and AI factories to agentic AI and inference, GTC 2026 will showcase the breakthrough shaping every industry. Learn more and register now at nvidia.com slash GTC. Today, we're diving into the future of public transit safety and efficiency. I'm excited to be joined by two guests who are at the forefront of this transformation. Ahsan Beggs, CTO of AC Transit, the third largest bus operator in California and main bus operator for the East Bay region of the San Francisco Bay Area.

0:53Shout out Alameda County. And Marty Beard, CEO of Hayden AI, a San Francisco-based company that's using technology to make roads safer and public transit better. Asan, Marty, thank you both so much for taking the time to join the AI podcast. Welcome. Thank you. Thanks for the opportunity. So maybe we can start with a little bit about what AC Transit does, and then we'll get to what Hayden does. And Asan, you can speak just a little bit about your role as CTO. And then as we go, we'll get into the collaboration that brought you guys together and brought us here today. Yeah, sure. Thank you so much, Noah.

1:28Thanks for the opportunity. Asanbe, Chief Technology Officer, Alameda Contra Costa County's transit. So a lot of times people think AC is only Alameda. So we do serve in Alameda and Contra Costa County. So basically, we're a two-county public transit system. I just want to clarify also, we are bus only. As you mentioned, Noah, we're the third largest in the state, and we are the largest bus only in the Northern California region. Our pre-COVID daily ridership was about 200 ,000 people. It's roughly about, you know, when you look into on an annual basis, you're talking about 55 to 57 million riders on an annual basis.

2:12I mean, that's a fairly significant mobility domain area if you want to look from that perspective. Our mission at AC Transit is safe, secure, reliable, and sustainable public transit. AC Transit is really kind of unique that our board is an elected board. It's different from many other public transit agencies. There are only possibly, I think, three public transit agencies in the country that have elected boards. Oh, is that right? What it really means, Noah, is that basically these are the board members, elected board members, are sort of really the people who are passionate about public transit and about mobility services.

2:53So I'm really proud of the team. I've been here for almost eight years managing the technology programs, innovation program, and love my job. And really, I'm passionate about providing those services to my customers in the East Bay. Well, as one of your East Bay customers, we appreciate you helping us get around safely and quickly. Marty, tell us a little bit about Hayden. Sure, yeah. It's great to be on. So, Hayden is a transit company that's focused heavily on AI technology to try to improve transit. So, our mission every day is really bringing together all the technology required to work with folks like us on and really try to improve public transit.

3:37And by that, we mean, you know, trying to help buses move faster, trying to reduce collisions, trying to make it a lot more safe for people that need help getting on the bus, etc. So we're a San Francisco-based AI company. We're experts in AI, but really we're experts in transit and the system of technology that you need to bring together to help do some of the things that I'm sure we'll get into. How long has Hayden AI been around? The company was formed in 2019. Okay, a lifetime in the current AI industry. Yeah, that is right, exactly. And, you know, at this point, we're on over 2 ,100 vehicles nationally and working in, you know, 10 major cities across the country.

4:21And we're also expanding internationally as well. Excellent. We've been doing this for a while. So how did Hayden AI and AC Transit come together? Did it start from a problem AC Transit needed to solve? What was the genesis of the collaboration? Yeah, I guess I can jump in, Noah. So, you know, as a part of my job is always looking for innovative solutions and technology that can solve some of our business problems, bring efficiency, improve safety, improve reliability. And I'm, of course, as a technologist, I'm a firm believer that if you have the right technology and you're attacking on the right business problem, you can make it happen.

5:00And, you know, we always hear and talk about people, process technology. Technology is part of the, of course, this whole solution. So, yeah, we have been looking into redeploying our dedicated bus lane system, which we call BRT, bus rapid transit connecting Oakland to San Leandro. So there's a dedicated bus lane. One of the challenges we had is that, you know, we had always illegally parked cars in those dedicated lanes. And we have been using legacy technology where it was requiring our operators to press a button to take the picture of illegally parked car in a dedicated bus lane. And the whole manual process, downloading the video, taking it to the sheriff's office, sheriff is reviewing the video.

5:48And typically, it really was creating sort of a stress for our operators because our success rate was less than 5%. So you're capturing all these videos and images, but your success from the citation perspective was less than 5%. So that was a major business problem. It's losing the effectiveness. We had the legislation. So we work with many different transit partners and we went to the legislative in the state. And basically, we work with our partners in crafting the new legislation, which is AB917, that authorizes us to use the automated lane enforcement technology. And, you know, one of the interesting things we did were not only we enabled this legislation to deploy this automation technology, leveraging AI, not only for the dedicated bus lane, but also the bus stops.

6:46So that's where, you know, I was looking for a solution. You know, I found about Hayden. I said, this is exactly what I'm looking for. So how can we work together? And, you know, we started the whole journey, starting with five buses and a pilot. And during the whole process, you know, we found, I mean, they are the best, you know, partner and the solution provider at the time. So we decided to move forward. And then we went to the board and got all the approvals. And now we have been working for almost more than two years now. Okay. And so when you first started working together, was the idea originally to use camera-based systems?

7:21And, well, stop there. Was that the original idea? That is true. Okay. And then when you first started deploying them, what were some of the early challenges? This is, I guess, a couple of years ago now, but what were some of the early challenges you had to get past putting the camera-based systems on public transit vehicles? I think the major challenge was, of course, making sure that it does the job with accuracy. Accuracy requirement was more than 90%. I was not getting the same accuracy with my legacy system. So that's number one. Our success rate was far less. I mean, as I said, less than 5%.

7:57So we were looking into image quality, lighting conditions, angle. I mean, the typical things you look for when you're looking for the lane enforcement and camera technology and computer vision and leveraging the AI. And the entire end-to-end from the time to capture illegally parked car from the dedicated lane to the bus stop or the bus stop. So at the end, the sheriff's office reviewing the citation and issuing the citation, we were looking for the entire process to be automated, not manned. And of course, improving some of those key performance indicators, what I mentioned. And the other thing we were looking into is the privacy.

8:37We wanted to make sure that the privacy is part of the whole design. So we are not capturing information just arbitrarily and keeping it. But that, of course, you know, we went through the whole privacy design criteria, making sure that no data, you know, stays on our system or our edge, which is inside the buses. So, yeah, so those are some of the key, I guess, success factors, you know, we defined now during the initial launch. And, of course, you know, it's not only just a technology, right? It's, of course, maintenance, operations, educating our operators, you know, sharing with our riders what we're trying to do, showing the benefits.

9:13So it was a pretty good whole process that took some time. But before we dig further into the process, Marty, can you talk a little bit about how this system works? Yeah, sure. Yeah, I mean, it's composed of hardware, software, and let's call it implementation services. So at super high level, the hardware are cameras. And as Hassan mentioned, the camera go on the inside of the bus. So it's literally on the inside looking through the windshield out into the right, into the bus line or the bus stop area, the curb. And so those cameras are optimized exactly for the use case that he just described.

9:50That then feeds into a, think of it as a control box that is not that big, that's inside the bus. That's where the magic happens. That's where the AI algorithms, and obviously it's all running on NVIDIA and we're huge, huge fans of NVIDIA and leverage NVIDIA's edge products a lot. Appreciate it. Anyway, you've got this control. That's where the AI is running. Right. And that's looking for the violation, right? That's optimized for that. And that's, Asan mentioned, that's running. It's an edge-based system. It's all edge. It's literally mobile, right? It's inside the bus. The bus is moving. The camera's looking.

10:30It sees a car that's, you know, blocking a bus lane or maybe blocking a bus stop. That is captured. So that image is captured and, quote, processed. And by that, it means the algorithm says, is that a car where it shouldn't be? Has it been there longer than it should be? Now I need to kind of package that video and package that information and send that to the right place to actually be reviewed and ultimately an enforcement sent out. So that's really it. The fancy term is called sensor fusion, which is really computer vision that's just looking for objects, but also location. So you need to be very, very clear about where a car is, when it's there.

11:09And you need to be very precise. I mean, this is obviously, we're trying to change behavior, which is we, ideally, we don't see any cars, right, in the bus lane, right? So it's got to be precise. It's got to be accurate. But those are the parts. It's the hardware, the cameras. it's the control box, the AI apps is a way to think about it. And then packaging all that in a way that's very private, very secure, and then sending it out to be processed. Right. And at the end of the workflow, when it's processed, does it go as far as deciding whether or not to issue a citation and then issuing the citation automatically?

11:46Yeah, we, and then Asan can take this as well, but we package what we believe is a violation, right, based on all the evidence and everything that we've pulled together. But that then does get sent for kind of ultimate review by somebody to say, yeah, we agree, and now a citation could be sent out. Got it. But we're only sending out what we believe to be highly accurate, captured enforcement. Hassan, you mentioned a moment ago talking about educating the public, educating everybody on the use of these systems. How has that been going? How have the drivers and the operators responded? How has the public responded so far to DePlanes automated systems?

12:28I mean, from the operator's perspective, Noah, of course, it's a big blessing that they don't keep pressing the button. And, you know, I mean, one of the things we always try to do is, as I mentioned, you know, safety is the core principle we follow and we adopt and we promote for our riders and for our operators, for our employees. So for operators to continuously monitoring whenever they're driving, but also paying attention to these illegally parked cars and making sure when to press the button and when not to press the button, what are the lighting conditions and things like those, some of those details.

13:06Now, this whole implementation has taken that whole responsibility away because everything is now pretty much automatic. Yeah. So operators' feedback has been very positive. They like it. Now, I think the one thing which is very important from the writer's perspective is we are seeing improvement in the on-time performance. We are still collecting the data and we're still going through the whole sort of this 100 bus pilot project. So we still need to develop a lot of KPIs and working with Hayden very closely. But we are already seeing significant improvement from the first-time offender. we are seeing a reduction 70%.

13:47So we're not seeing those keep repeating. We are seeing improvements even in the on-time performance. We are seeing the improvement in the accessibility where an illegally parked car at the bus stop was blocking our bus to park and enabling our accessible needs rider to get on the bus. So a lot of those metrics and KPIs, we are in the process of mining a lot of this data comparing with our historical data, what was some of those challenges, and even seeing the accuracy, you know, from our sheriff's office perspective because of the sole automation, what, you know, Marty was talking about. We are seeing an uptick in the improvement on the accuracy of the information.

14:30So I think all together, it's going into the right direction. That's great. Being a public agency, as you mentioned at the beginning, you know, the work that you do is subject to obviously following legislation, new legislation being passed, board approval, all of that. What do you think is important for the general public, the riders of the transit system, but also the policymakers who set these laws and rules to follow? What's important for them to understand about using this kind of technology the way you are? Yeah, I can dive in. Well, I mean, most importantly, and given how much experience we have, it works, right?

15:06So the focus is on improving the transit rider experience. At the end of the day, that's the customer, right? And we see that. So if buses are moving faster through a network, that has a huge impact on people's lives, just in terms of on-time arrival, in terms of getting from point A to point B faster, et cetera. And then you get reduced collisions and you're increasing access and safety. So all those metrics that Asan mentioned, we track those religiously and it works. That's what motivates us, right? You know, it kind of, it works. I think the second thing is, when you talk about AI and cameras, I mean, people immediately just back up and go, okay, that's creepy.

15:47And it's kind of like, okay, yeah, just step back for a second. This is not looking at people. There are no people identified, right? This is only looking at vehicles and only vehicles that are where they shouldn't be, right? And at the end of the day, and so I think we have to educate sometimes, like, look, even if somebody asked me for information about identity, I don't have that. We don't keep that. Nothing's stored. Hayden doesn't have that. I don't have that, right? So all I have is I have the vehicle. I have the enforcement criteria that was given to us. And so I think we have to educate on that, just given—and I understand, right?

16:24I mean, I get it. It's an emotional issue around privacy and so forth. It's complex, sure. Yeah, it's complex, and it should be, and we should think deeply about it. But I think in this case, it's very use case specific, what we're talking about. And it works, right? So, yeah, I mean, I think, yeah, that's our responsibility. But, you know, whenever we are adopting new technology or new tool, you know, we need to make sure as a public entity, public organization, that, you know, we have ample education, knowledge sharing, information sharing. And we do this through, of course, our legislative process.

16:59So, you know, when we decided to move forward after the whole, you know, request for information, looking at the entire industry, who can provide really those specific elements of what AC Transit was looking for, we checked the market. We published the whole request for information. We got, you know, proposals. As a result of the whole evaluation process we went through, we decided to go move forward with this, you know, this specific technology from Hayden. And we took it to the board and we educated them, we presented to them some of the things Marty was talking about specific to privacy. And we wanted to make sure that we are in compliance with our local privacy policies and not capturing any information about faces of our users or writers or people.

17:47This is all forward-facing. This is all about license plate and only under certain conditions, parameters defined by us. And even within that, it's specific, like, for example, bus stop. You know, we implemented these bus stops as almost like a digital twin in the system. So if someone is parking and my bus is approaching, that's like not good for our buses to demonstrate the on-time performance. So we captured this information. And the same thing with the state legislation. You know, if you look into the AB 917 that was adopted by the state of California, our assembly and our Senate, and then eventually signed by our governor, it was kind of a fairly rigorous process to demonstrate with the data that, you know, how is it going to be helpful?

18:36And, you know, I'm really proud of our legislative team. They've worked extensively and they're still, you know, asking the information and the data for us to provide because this existing legislation is set to expire in 2027. So we have to continuously demonstrate the value of this technology and provide this information in a very specific form, what they're looking for so that they can educate public and we can educate our lawmakers, policymakers. makers. I'm speaking with Asan Beg and Marty Beard. Asan is the chief technology officer at AC Transit, California's third largest operator of buses and the largest operator in the East Bay region of the San Francisco Bay Area, where I call home.

19:22And Marty is CEO of Hayden AI, a San Francisco-based technology company that's been working in AI and transit, improving efficiency and safety for public transit systems and riders for the better part of a decade now. I want to sort of take a step back and look at the broader picture. And Asan, maybe I'll start with you. From your perspective as a CTO, we talked a little bit about the specific problems that you were looking to solve and how you started working with Hayden. How does a project like this fit into your broader purview as CTO of AC Transit and kind of the the bigger picture for AC Trends' roadmap, if you will, for digital transformation and, you know, embracing technology, leading-edge technology?

20:07Yeah, I mean, that's a great question, Noah. So, as I mentioned earlier about Chief Technology Officer, I talk about the elected board. So, that's sort of like an advantage, you know, I guess we have as a technology practitioner that our board, you know, firmly believes in technology as a core integral part of the services delivery, what we do. And, you know, my general manager, my boss, you know, is a firm believer in technology as well. So we always found that, you know, they're very, very supportive for these kinds of initiatives. So, you know, when this whole issue came up about looking into really modern AI-centric technology, you know, we went through the whole step-by-step process, which is, you know, conducting the whole POC, five buses, demonstrating the value and looking into technology, looking into security, looking into cybersecurity.

20:59I mean, all multidimensional evaluation, but more importantly, tying to the business. You know, I don't believe in deploying any technology or a technical solution if it is not solving my business problem. So I basically partner with my chief operating officer at the time, CIO, I mean, CTO and CO coming together and trying to solve this problem with a vendor partner like Hayden, that was really sort of a success. So yeah, so in broader, I guess, spectrum, we always look for these opportunities where we can find cutting edge technology may not be fully proven, but sometimes you have to take those kinds of risks.

21:42So I think we believed in it. We saw that, the value. And, you know, we went through the whole process and really, I think it's been working out pretty good so far. Marty, from Hayden's perspective, you can comment on the AC transit relationship and specifically if you like, but also what else are you seeing? What else are you working on? How do you see AI shaping, you know, public transit and transit kind of more broadly? Yeah, no, huge. Well, I mean, we're able to work with innovative folks like Hassan and his team, and that helps a ton. Because I like this comment that at the end of the day, I could talk about some ethereal strategy about AI, but it's really just, can we practically apply it to help cities perform better?

22:25In this case, we're focused on transit. And so that easily extends into bike lanes. Can you help, can you, you know, can you manage, help manage bike lanes and try to get people feeling safer and able to leverage biking? What about parking more generally? What about other assets in a city? So recently we've been working on what's called roadworks identification where construction zones, right, have a massive impact in a city the size of like an Oakland or like a New York or something where, you know, it just has a huge impact on people getting from point A to point B. For sure. Can these cameras identify accurately a construction zone?

23:04Is it permitted? Did they get the permit? Not the permit? So these type things are starting to kind of logically come up because we sort of have this mobile AI going through an urban environment and capturing more and more information, right? So it's very logical extensions of what we do. We talk a lot up here about practical AI, practical AI. I hear people come up with expressions like cognitive cities and things like, it's like, I don't know what that means, right? But I do know that we can help manage assets. You know, we can help transit, we can help buses, we can help bikes, we can help parking vehicles, you know, et cetera.

23:41So that's when I look out, it's kind of like practically extending where it makes sense and adds value ultimately for the city managers. If we could collab and rig up some kind kind of a pothole filler that we could attach to the back of the AC buses. I'd go a long way in my neighborhood right now, but that's a separate conversation. As a cyclist, I would definitely agree with that. Yeah, right, right. Asan, public agencies, transit agencies often operate under, you know, just more restrictions, more constrictions than, say, a startup or a privately funded agency might. You have budget, procurement, policy constraints, and boards to deal with.

24:18And as you said, it's an advantage, but also things that you have to cope with. What advice would you give to other public agencies, you know, your counterpart CTO at a public agency somewhere else, considering tech initiatives like this? Yeah, I think definitely that's a challenge, you know, you're right, working in the public sector. But, you know, I mean, I think I found that public transit is really at the crossroad where we have the big responsibility to provide the mobility services. And technology is playing a very critical role in providing those mobility services. Whether you're talking about the longest stretch, you're talking about the middle mile, or even if you're talking about the last mile.

25:01My advice is to, you know, really to always focus on the business problem. You know, what is the challenge? What is the issue? What is the core mission? You know, I'm trying to continue making it happen with the technology and with the technical solution. I guess I'm lucky that I'm in the valley, in Silicon Valley. And, you know, I find companies like Hayden, you know, startup. And I think I have, as I said, you know, I'm lucky that I have, you know, this wonderful board and great executive team that they believe in technology and they believe in trials and POCs and pilots to really, you know, fail fast, you know, sort of a strategy that, you know, you need to try.

25:43And you need to see what is going to stick and what is going to work under certain criteria. So I've been lucky. I think there are lots of opportunities. There are lots of national organizations. And public transit is kind of a really, you know, very well-connected community. And one good thing about public transit and public sector is nothing proprietary, nothing, you know, intellectual property that I'm holding. So if I have success, if I have good methodology, good finding, and a way to make it happen, you know, we all share. So I think just be bold, just to try out, you know, and really shoulder to shoulder with the business.

26:21I think that, to me, is the most important thing. Marty, learnings from Hayden's side, either that could be applied to, you know, someone in a public agency somewhere else, or to other, you know, practitioners using AI to try to solve transit problems. I mean, yeah, I think Asan said it really well, which is, what's the problem, right? What is the business problem you're trying to solve? I mean, coolest thing about public transit is we're talking about thousands and thousands and thousands of vehicles providing millions and millions and millions of trips, right? It has a massive impact on our country and our states and our cities.

26:58So I love being in the middle of like, okay, what's the biggest challenge that we're facing here? And how can technology, whether it's AI or machine learning or whatever you want to call it, how can it help? And the cool part is it can, right? So I think it's fun to go out and kind of, quote, sell the vision because you know it can work. So you kind of come in with confidence and you're sort of like, let me show you some data and let me show you some real activity. So I think versus being in a lab and working on AI just kind of ethereally and sort of thinking through it, it's fun to be out in a physical space like a bus or, you know, and kind of like, okay, what can we do here to try to add value?