![Podcast microphone against a blank background](/sites/g/files/omnuum10826/files/styles/hwp_21_9__1920x825/public/datasmart/files/podcast_microphone.jpg?itok=3Q9qYvou) # From Reactive to Preventive: How AI Transforms Public Works Episode Ninety-One April 01, 2026 [ Betsy Gardner ](/people/betsy-gardner) Most cities respond to infrastructure problems after residents report them. What if they could detect and prevent them first, while serving every neighborhood fairly? Host Stephen Goldsmith sits down with Daniel Pelaez (CEO of CYVL), Khahlil Louisy (Public Innovation Institute), and Mike Dennehy (former Boston Public Works Commissioner) to explore how artificial intelligence and computer vision are revolutionizing infrastructure management, closing equity gaps, and helping cities shift from reactive operations to predictive maintenance. **In this episode, you'll learn:** - How computer vision detects infrastructure problems before citizens report them - Why traditional complaint-based systems can miss concerns in lower-income neighborhoods - How natural language queries democratize access to infrastructure data for city managers - Why a "multi-modal" approach combining AI, citizen input, and external data delivers better equity outcomes - What cities can expect from predictive infrastructure systems Embed *Listen here, or wherever you get your podcasts. The following is a transcript of the conversation.* **Stephen Goldsmith:** This is Stephen Goldsmith, professor of urban policy at the Bloomberg Center for Cities at Harvard University with another episode of our podcast, Data-Smart City Pod. And we have a full house today. I'm going to introduce each guest briefly. We have Daniel Pelaez, who is the CEO and co-founder of a company called CYVL. Khahlil Louisy, who has been associated with us at the Harvard Bloomberg Center, is executive director of the Public Innovation Institute. And Mike Dennehy, who is former commissioner of the City of Boston Public Works. Welcome to all three of you. Why don't you each give our listeners just a minute about your impressive backgrounds, and maybe you can even pronounce your last names better than I did. Let's start with Daniel. **Daniel Pelaez:** Thanks, Steve. Really awesome to be here with quite the crew. Yeah, I'm one of the founders of CYVL. Quick background origin story on us. Our company, we help cities better manage their physical infrastructure. First job I had was working on the road crew of a Public Works Department in Southbury, Connecticut when I was 18 years old. And I just saw firsthand that these guys had no data to manage roads, sidewalks, or really any other asset. They were collecting this data by hand and it was living on paper records. I studied self-driving cars and robotics in undergrad, went to school in Worcester, Massachusetts at WPI. And that's when I sort of came up with the idea of taking awesome technologies from the self-driving car space and applying it to Public Works to help our cities build infrastructure better, manage it better, just get the job done with fewer resources. That's what we're all about, working with about a third of the municipalities in the Commonwealth of Massachusetts, and about 500 towns or cities across the US. **Stephen Goldsmith:** Mike? **Mike Dennehy:** Yeah, thanks, Steve. Mike Dennehy, former Public Works commissioner, city of Boston. I spent 20 years in the city. I started as a budget analyst and was immediately attracted to operations. Spent most of my years there, including Mayor Walsh's first term. I then went on to be the town administrator in Milton, Massachusetts, and joined the CYVL team back in January. **Stephen Goldsmith:** Good. Glad to have you, particularly in the airport or wherever you're hanging out right now. Khahlil? **Khahlil Louisy:** Hi, my name is Khahlil Louisy. I am an economist by training, focused on the fields of public economics, health economics, and economic development. And in particular, looking at the role of technology and innovation across each of these domains. For the last five years or so, a lot of my work has been focused on health, health systems, the application of technology and innovation to improving health outcomes at the population level. And most recently, a lot of that work has been focused at the city level, and we've expanded scope. Realizing the limitations of innovation at municipal levels, I founded, with a small team, the Public Innovation Institute, where we focus on applied research, innovation, development of technologies to improving population outcomes. **Stephen Goldsmith:** Thanks. Khahlil, let me start with you, and I want to focus on responsiveness and representation. You have a [recent paper that looks at how residents and algorithms see problems differently](https://datasmart.hks.harvard.edu/residents-algorithms-see-different-problems) using Boston 311. If the purpose of government is to improve the quality of neighborhood life, at least the way government works today, it responds to a problem. Tell us a little bit about your paper and what you found in terms of comparing algorithmic versus citizen initiated requests. **Khahlil Louisy:** Yeah. Well, the first thing to point out is that many cities are now resorting to technology to better understand the problems that their residents are encountering on the daily. 311 historically has been one of the mechanisms used for citizens to, first, report challenges in their communities and then for government to respond. Recently, cities are resorting to the use of AI, AI-enabled systems and CYVL, their technology, computer vision system with AI, is one such technology that is used to conduct the level of monitoring that cities perhaps wouldn't be able to do. They wouldn't have the capacity or the capability to undertake regularly. And so, I thought it might be an interesting case to use this technology to understand, to compare and contrast the difference between the 311 reports from residents and the use of the technology. Are they picking up the same things? Are we addressing the equity issues by using this technology? And the findings were, in some respect, fascinating, but others I think intuitively we already knew. One is we found that communities, or neighborhoods that are highly engaged with government, had a higher reporting rate with 311. Governments typically allocated greater resources, a greater share of resources to those neighborhoods. Versus neighborhoods that were lower income or lower standing on economic distribution. And those were the minority areas, large immigrant populations. Another thing we found was that engagement, were lower in some of those areas, the technology was picking up things that people ordinarily wouldn't resort to. You're able to pick up details as more as hairline fractures or pavement distresses before they become larger problems that resort to personal injury. So the technology's able to pick it up. **Stephen Goldsmith:** Let me move to Daniel for a second. Daniel, a little bit about the point that Khahlil's making. How, in today's world, can visual analytics and sensors and the processes such as the ones you use, how can they anticipate a problem and therefore alert the city and not necessarily require a citizen complaint to initiate it? **Daniel Pelaez:** Yeah. If you even look at current reporting standards for infrastructure, whether it's for municipalities, state DOTs or federal governments, they're written around a manual way of doing things. When I say manual, like some person having to go out with a clipboard and a ruler. And because it's literally impossible to have an army of people meticulously with a trained eye of a civil engineer constantly inspecting infrastructure, these standards or recommendations are really built on ‘try to inspect your infrastructure every other year.’ And even for municipalities, the guidance is do 10% of your roadway network, try to extrapolate out because it will be impossible to catch everything. And the goal of that is people are building infrastructure asset management plans, and they're trying to forecast and predict, ‘what is this asset going to look like six months from now, 12 months from now? How are we building our capital plan around these major infrastructure investments?’ But everyone we speak with, every city leader knows that those forecasts, it's not a crystal ball. Things change. It's not going to be that accurate. Something's going to pop up six months from now that it goes unnoticed, turns into a $10 million issue, and your capital plan is completely out of whack. These sensors that we're using, both cameras and 3D mapping technology, we can take these same sensors that are inspecting the entire world for these other applications and apply it in the way to inspect physical infrastructure. And since it's an automated approach, it's collecting the same data every time. There's no human subjectivity to it. It's able to produce very consistent, repeatable results. So they can identify that hairline fracture and how much it's maybe widening over time so they can proactively go fix that thing before it becomes an issue. **Stephen Goldsmith:** This is quite interesting. We've been doing a lot of work on community engagement. Tell us what the difference is between an LLM and an LVM. And could we enlist a group of neighborhood volunteers just to run around with their iPhones and take pictures to augment this as well? Could that process be used in conjunction with what you've discussed? **Daniel Pelaez:** That is an amazing question. LLMs and VLM, there's large language models, and then there's video, video learning and video language models. This is going to get a little technical. I'm going to nerd out a little bit for some folks. What we are doing is we're taking sensors, we're mapping streets, we're taking this very heavy camera imagery, and we're essentially indexing everything. So if you think about how Google indexed every single website ever, they could figure out what is in that website and then you can search it very quickly. Our really core technology is being able to do that. Take all these pictures, figure out very quickly what's in it, whether it's a pavement crack or something wrong with the sidewalk, and then make it very easy for another AI system to access that data. We can take in other data sources, other forms of picture imagery, video imagery, satellite imagery, and apply all this to what we're calling the index of the physical world, or an infrastructure index, and then enable literally unlimited amount of use cases for a resident or a city leader to query this data, find information quickly. So to answer the question simply, yes, someone with natural language could say, "Show me the dangerous intersections based on our prior data that's in our dataset with maybe faded pavement markings and high traffic." I know I went down into a rabbit hole, but it's possible. **Stephen Goldsmith:** That was a pretty good rabbit hole. We like that rabbit hole. Mike, lift us out just a little bit. I know that you and Daniel work together, but let's go back to your last job. Boston's very advanced in its use of AI. If you had these tools, and if you could teach your deputy commissioners of Public Works to make natural language inquiries of the data that Daniel said, how could we operate smarter? How could you operate a department that's more preemptive, that's more predictive, that's more anticipatory, based on what Daniel and Khahlil said? **Mike Dennehy:** Yeah, Stephen, that's a great lead in. What Khahlil was onto is very relevant. We started using data in the early 2010s. The administration built a really engaging platform that Khahlil had talked about, rolled up into a really neat daily dashboard. But on our end, it was used to move personnel around, literally just reprioritize your day-to-day operations. It helped us do more with less, but it had very little impact on our budget and our ability to maximize how efficient we were with the tax dollars when it came to planning and making sure that we addressed the infrastructure in a timely manner. If potholes were the priority, litter baskets would have to remain overflowing. There was no planning element built into this data, we simply used it for day-to-day operations. Don't get me wrong. We became a better operation, but we were so reactive. It was literally a daily shuffle. And our walk downstairs from the seventh floor to council chambers on the fifth floor was ultimately filled with uncertainty and a general hope that the body would accept another year of our static and basically historic requests. This technology is a game changer. It would allow these departments to analyze and process data so quickly and precisely that go from reactive to proactive very, very quickly. To Daniel's point, it helps you build out your CIPs. And again, I wish I had this intelligence in my arms when we walked downstairs to council chambers back in the day. My advice would be to the deputies is just you have to put your arms around this stuff. It's really going to help you build and maintain infrastructure. During that unfortunate time in office as the commissioner, we had to shut down the Long Island Bridge. And when I say Long Island, I'm referring to the Long Island Bridge that connects Quincy to one of the Boston Harbor islands; dilapidated because of years of not looking at the infrastructure. This technology allows you to address it in a very timely and financially responsible manner. **Stephen Goldsmith:** One more quick question to you before I go back to Khahlil. How far are we away from a point where the middle management of Public Works could make a natural language inquiry of spatially layered data like, ‘show me repetitive potholes, show me the repetitive potholes as contrasted to or augmented by drainage issues or tree issues?’ In other words, how close are we to being able to democratize the availability of the data and put it into use? **Mike Dennehy:** We are there. It's just a matter of convincing those middle managers to convince the budgetary and sometimes elected officials that this is the course of action to take. Daniel can extrapolate on that. But literally, with those tools in place, it's obligatory for these middle managers to then make this the priority. **Stephen Goldsmith:** Daniel, since Mike just passed the baton to you, give us some examples of how to translate what you do and the conversation we've had into a better quality of community services. **Daniel Pelaez:** Yeah. I'll give some examples as to what we're doing, even some research projects with the city of Boston. Santi Garces, CIO, probably the most forward-thinking CIO I've met for a city, they're working on their own internal agents. You can ask this thing anything, and it's going to tell you exactly what's going on and what to do. The connection we built is to essentially let you talk to infrastructure. We built an MCP for Boston's infrastructure. Santi and his team can now go into this thing and just natural language say, "Hey, there's some flooding going on this intersection at this address. Help me find all the manhole covers and the catch basins. Show me the geometries of the roadway and overlay that with resident complaints. Give me a report I can give to the mayor so she's going to get a really quick readout as to what's going on." **Stephen Goldsmith:** Khahlil, let's take all of this, the conversation and the technology that's been described. Design for us a system of responsiveness that's more equitable. **Khahlil Louisy:** Wow…here's what I think. I think the technology presents a fantastic opportunity for leveling the equity playing field. We do have to be mindful of a few things and take that into consideration as we're designing some of these systems. One is we have to ensure that we're benchmarking the systems against established ground truth. As Daniel was saying earlier, two different surveyors can go to the same spot and come to different conclusions, right? And if you're using that information to train the technology, depending on which surveyor was closer to the truth, the technology itself might give you an answer that somewhat exacerbates the equity issue. We have to figure out, how do we establish what is truth? How do we develop the standards for it and benchmark it and use that to train the technology? We cannot just use the technology alone, so that's the other part. It has to be a combination or multimodal approach to designing some of these systems. We use the technology in combination with. And that's because some of the residents may not be as connected, they may live where the technology itself doesn't pick up. Some of their priorities are different. It's a multimodal system. We use a combination of citizen reports and we use the technology, and then we use external data to understand what is happening to establish ground truth and then to design a system that improves equity. **Stephen Goldsmith:** Daniel, two questions for you. One is, cities are having a difficult time procuring solutions from companies like yours. And for the following reasons: they're not able to understand enough the obsolescence trajectory, right, so should a city buy a solution or do they purchase certain services? How do you think about what the right model is for the city, A? And B, what does this business look like 12 months from now? **Daniel Pelaez:** Procurement certainly is a challenge. And when you add a brand new technology where people aren't really sure, ‘is this software, is this data, is this services?’ It becomes a little complicated. When the agency understands the value and you have a champion within that is able to explain to Procurement and Finance why this is so important and why the ROI is going to be so massive for taxpayers, we're able to find a way. Fortunately, the cost of this technology is far less than what it takes to do this stuff by hand. And the last thing is partnerships. We've been able to develop this ecosystem, a partner ecosystem will work directly with the office of the CIO, then they can help from the top-down, figure out how all their agencies within, not just Streets, but Planning, Parks and Rec, Public Safety all leverage this digital layer. Now your last question, 12 months from now, infrastructure intelligence, I think it's going to start to become commonplace within governments. It's been very exciting, the traction we've seen, just over the last six months, really. So, 12 months from now, I think we'll have unlocked some things for agencies we wouldn't have thought of before. And I think some state governments are going to see network effects as to what happens when half of the state is using the same technology, we're able to have the same insights at much larger scale. I think this is the next big thing after LLMs. **Stephen Goldsmith:** It was helpful. Mike, let's say you went back to public service and you led a major city Department of Public Works, say six months to 18 months. What does that department look like? How does it differ than it did a year ago? How will it turn all of this data into action? **Mike Dennehy:** Yeah, that's a great question. What I see is middle managers walking into budget hearings with folders that include intelligence reports that are being generated by this infrastructure intelligent platform that specifically call out things that may not have come into 311, right? Things that this technology has addressed and can provide preventive maintenance or more long-lasting effects from a repair basis. And when you talk about the degradation curve, something that we don't address now could be five, nine, 10X, three or four years from now. I'm very parochial when I speak about my government days. There are other entities within city halls and town halls, Transportation Departments, Parks and Recreation Departments, anybody that owns infrastructure within the public realm. Addressing those issues in a more timely manner will find this intelligence to be very worthwhile. **Khahlil Louisy:** If I can add one thing quickly, I think this data is very rich for informing decisions across sectors. We looked very specifically at health, how this infrastructure data can be transformed into epidemiological signals and assessing environmental risk factors. But I think there are other areas that we've been probing into, like commerce. How can this data be used to inform where within a city might be the most viable place for commerce, for commercial locations? And then there are other sectors – environment is something that we talk about, and I know that Data-Smart is particularly interested in, but other areas including education and finance. We should be looking into it, and I think there are opportunities for it. The other thing is, I think we should probably try to estimate the economic benefit of using this technology. How much are we actually saving cities by utilizing this technology rather than having people manually go on each street and not actually able to physically meet every single street within a city? **Stephen Goldsmith:** A great set of observations about return on investment and equity, and an excellent conversation. My thanks to Daniel, Khahlil, and Mike for helping us focus on the potential opportunities of data neighborhood responses in Public Works. This is Steve Goldsmith from Data-Smart City Pod. Thanks for your time today. ## About the Author ### Betsy Gardner ![Headshot of Betsy Gardner](/sites/g/files/omnuum10826/files/styles/hwp_1_1__100x100_scale/public/2025-05/Betsy%20Headshot%20resize.jpg?itok=k2OsSp1g) Betsy Gardner is the editor of Data-Smart City Solutions and the producer of the Data-Smart City Pod. Prior to this, Betsy worked in a variety of roles in higher education, focusing on deconstructing racial and gender inequality through research, writing, and facilitation. She also researched government spending and transparency at the Lincoln Institute of Land Policy. Betsy holds a master’s degree in Urban and Regional Policy from Northeastern University, a bachelor’s degree in Art History from Boston University, and a graduate certificate in Digital Storytelling from the Harvard Extension School. See also:- [ 311 ](/topics/311) - [ Artificial Intelligence ](/topics/artificial-intelligence) - [ Equity ](/topics/equity) - [ Infrastructure ](/topics/infrastructure) - [ Operations ](/topics/operations) - [ Predictive Analytics ](/topics/predictive-analytics) Share on:- [ Facebook ](#) - [ Twitter ](#) - [ Linkedin ](#)