Using Artificial Intelligence to Find Areas with Potential for Housing Development in New York City
Advisor: Mr. Rappelfeld
Internship Location: Rafyana Realty Corps
Onsite Mentor: Sam Shpilfogel
Urban development and investment are often uneven, which can leave certain neighborhoods overlooked for decades while nearby neighborhoods grow quickly. My project breaks down how Ai could identify these underdeveloped urban areas to find opportunities for responsible real estate investment and community growth. At the same time, I will look at the human factors that explain why some areas remain untouched. I want to see if AI tools can analyze public data, such as property records, development history, and demographic trends, to highlight neighborhoods that have seen limited development over the last 25 years.
I am conducting this research in partnership with Rafyana Realty Corps under the guidance of Sam Shpilfogel, but I still will be in contact with his team. The project involves gathering human and real estate data and using AI to analyze patterns of development across different neighborhoods. My models will look at factors like construction activity, property values, zoning changes, and infrastructure investment to identify potential development opportunities. I will be asking for advice from Sam and I will try to mimic their strategies using Ai to improve and help the strategies and patterns AI will use and build.
The goal of this project is to show how AI can be used as a tool for real estate professionals to help find overlooked areas more efficiently. By making it easier to discover investment opportunities. This AI assisted analysis could help lead to smarter urban planning and encourage growth in communities that have not yet benefited from recent economic trends.
Project Blog
Week 0: Introduction
Hi, my name is David Gladkovitser and I'm very passionate about real estate. I have been at BASIS Independent Brooklyn since the 6th grade and I have developed a strong interest in business, investment, and seeing how cities grow and change.
My Project is about using Artificial Intelligence to identify High-Impact opportunity zones in Brownsville, Canarsie, and Sunset Park. Specifically, it looks at long-term development patterns using historical data like permits, zoning changes, construction records, and demographic shifts to identify areas with little to no development. This leads me to my goal of simplifying the process of finding these Opportunity Zones with solid future potential.
Instead of relying only on human judgment, I will be using ChatGPT to analyze large amounts of data and identify patterns that might not be obvious at first. At the same time, I also want to understand the limitations of AI and how it compares to real-world experience.
My placement at a real estate office not only supports my academic research, but also gives me real industry experience. I will be going there multiple times a week to observe, ask questions, and learn directly from professionals. This hands-on experience will help me connect what I learn in my project to how things actually work in the real world. Overall, I am excited to start this project and explore how AI can be used as a tool in real estate, especially as I plan to pursue this field in the future.
Week 1: Starting My Project
My senior project started this week, and I wanted to spend the first few days really getting familiar with what I’m actually doing. My project focuses on using AI to help analyze real estate investment opportunities, specifically in low-income neighborhoods. I’m working with a real estate office where I’ll be meeting a few times a week with experienced professionals to learn the strategies they currently use and work with real data.
At the beginning of the week I spent time thinking about how opportunity zones are evaluated. Things like location, price trends, demand, and timing all play a major role and can completely change the value of an investment. I also started thinking about how AI could speed up that process by analyzing large amounts of data much faster than a person could, especially when comparing multiple areas at once.
One thing my on-site mentor emphasized, and challenged me on, was just how much data is involved in real estate. It’s not just about one property, but entire neighborhoods, different patterns, and long-term trends. That made me realize that I need to stay focused on specific zones so I don’t get overwhelmed.
Overall, I think I had a strong start. I have a clear vision of what I want to do, which is to use ChatGPT in a targeted way by developing and testing different prompts to identify potential opportunity zones. Over the next few weeks, I want to start working more directly with the data and begin running tests to see which AI prompts are actually effective.
Week 2: Cracking down the Basics in Real Estate Data
This week I started to understand real estate data much more as a real estate professional. In the office, I saw agents taking action, handling property listings, comparing prices, and determining if something is a good deal.
I learned about things like price per square foot, rental income potential, and neighborhood trends. These are all important when deciding if a property is worth investing in. I also started thinking about how I can use AI to analyze these factors faster.
At the same time, the market has been one thing that I didn’t expect to be so unpredictable and volatile. When evaluating deals, some opportunities may initially seem promising, but there is often a lot of uncertainty and risk involved. Even when something looks like a good deal on paper, there is always a question of how risky it actually is. This made me realize that AI will not replace human involvement in these decisions. It may not replace human decision-making, but it can definitely support it.
I also started to experiment with AI tools and asked simple questions like “What are some of the qualities of a good real estate investment?” and “What are other aspects of property value?” This helped me think about how to design stronger prompts in the future and how to improve them over time. Overall, I feel like I am making progress. I’m starting to connect the real-world work I see in the office with the technology side of my project.
Week 3: Building My AI Strategy for Real Estate Opportunity Zones
This week, I began introductory research in earnest on my project centered on the utilization of artificial intelligence in identifying real estate investment opportunities. In particular, my area of interest will focus on determining the best lots in Opportunity Zones in Brownsville, Canarsie, and Sunset Park on which to build. I believe there is a real chance here to build a model that will get a jump on the competition because AI technology is relatively new and rapidly evolving, and many real estate firms aren't studying these minute-by-minute advancements.
As I'm especially interested in areas that have not seen development in a long time, which might represent investment opportunities that have gone overlooked, I spent most of my time working on my methodology for collecting the data I plan to feed into an AI model designed to analyze patterns and evaluate possible investments. To do this, I began poring over available public records, permits, zoning, and census data. Principally, I collected data throughout Opportunity Zones in these three neighborhoods, where federal tax incentives are meant to encourage development in historically underinvested areas.
On Thursday, I spent time with Sam, the owner of the office I’ve been working with. What stood out to me in our conversations was how he stressed the human aspect that still prevails in real estate despite the increase in AI tools. There are always intangible factors to be taken into consideration, he pointed out, like neighborhood perception, informal knowledge, and even gut instinct. I'm going to have to figure out how to account for these as I iron out my approach.
I also discovered during data collection this week that the data itself is quite fragmented. It is much more inconsistent on a block-by-block basis than I hoped, even though I was well aware of this possibility, because permit records, zoning maps, and census data are often organized using inconsistent naming conventions and even geographic units.
Week 4: Walking the Grid
This week I got more hands-on, moving into the actual neighborhoods I’ve been trying to understand through the data. I went with my onsite mentor, Sam into several Opportunity Zones across the borough of Manhattan, partly to ground what I’ve been seeing in permits and zoning maps in something physical, and partly to make the project feel less abstract and more real and tangible.
What stood out immediately was how uneven development really looks when you’re standing in it. On paper, there’s a sense of similarity from block to block that quickly disappears once you’re on the street. That might seem obvious, but it was a useful reminder that I’m not doing this work in a vacuum.
We paid particular attention to empty lots and other transitional areas, and I immediately started noticing patterns I hadn’t picked up from the datasets alone, like proximity to transit entrances that are technically close but feel much farther, or buildings that seem viable except they’re surrounded by properties that clearly suppress their value. These are variables that don’t really show up in zoning or permit data but become obvious when you’re actually there, and they’re important when thinking about investment potential. All of this made me realize that my model, at least in its current form, risks flattening reality and reducing complexity, which isn’t my goal.
The biggest takeaway this week is that the trip helped me refine what I’m asking of the AI tools. Instead of just identifying underdeveloped lots based on time since the last permit or construction activity, I’m starting to think about layering in other indicators such as proximity to transit (and whether it’s actually reliable), frequency of local business turnover, the condition of neighboring properties, access to major roads, visible signs of investment, and crime rates. If I can build a model that reflects these variables as well, I think I’ll be in a stronger position.
Week 5: Sunset Park Analysis
This week, I decided to focus on one neighborhood: Sunset Park, Brooklyn. On paper, Sunset Park doesn’t immediately stand out in any particular way. It has some gentrification, but not much compared to other Brooklyn neighborhoods. Because of this, it is also marked by relatively stable property values. It hasn’t spiked like Williamsburg or even Greenpoint in recent years, for instance. At first glance, these facts might indicate a less attractive neighborhood for investment. But I decided to dig a little deeper.
What I ultimately found most intriguing about the neighborhood is that while it is somewhat underdeveloped in parts, it is also in close proximity to nearby, recent, and lucrative growth. Industry City, which sits at the southwestern edge of the neighborhood near the water, has proven that capital can flow into this area and generate significant returns on investment. The only thing is, this investment has been restricted to just a few blocks, it hasn’t expanded outward. Interestingly, though, the city is planning, in partnership with the Hudson Yards Development Corporation (HYDC), to renovate Bush Terminal and build out a new development further north along the river. This is also a very auspicious sign for things to come, even in residential areas that won’t be directly affected by the renovations.
Now that Industry City has been a proven success over the last half decade, I think it’s a good time for investors to reevaluate the rest of the neighborhood. After all, this kind of containment is unusual. Once a major anchor project like Industry City is built, and particularly if it succeeds, it is very likely that surrounding residential areas will begin to reprice relatively quickly.
Week 6: Block-by-block Strategies
This week, I spent my time at the office learning how to scout properties and devise strategies on a block-by-block basis. With Sunset Park as my continued target, I narrowed my attention to smaller patterns at the most local level, gaining a deeper, more nuanced understanding of real estate investment.
Near Industry City along 3rd Avenue, I focused on underutilized buildings positioned in commercial spillover zones. In quieter residential areas like 7th Avenue (between 44th and 50th Streets), I shifted toward long-term holds, identifying undervalued multifamily properties that could benefit from potential future expansion. I also looked for early signs of development, such as construction or renovation, new businesses opening, and exterior upgrades. This approach underscored the importance of thinking strategically about timing, not just location.
Another valuable tactic was identifying small inconsistencies, or “outlier properties.” For instance, even on more developed blocks, like those between 6th and 8th Avenues and 50th and 60th Streets, there were properties in poor condition. These findings often represented opportunities to add value in areas that are otherwise stable and relatively low-risk for investment. This showed me that the best opportunities are not always in underdeveloped areas, but also in the overlooked gaps within well-established blocks.
It has been interesting to see how investment strategies can vary greatly even within the same neighborhood. What’s even more compelling is spotting subtle behavioral changes, such as early shifts in tenant demographics or interior upgrades. These signals often reflect growing confidence in a block before visible development occurs. Investors who recognize these patterns can position themselves ahead of the market and capitalize on key investment windows by evaluating neighborhoods on a block-by-block basis.
Week 7: Reading Between the Blocks - Williamsburg, Brooklyn
This week, I shifted my focus to Williamsburg, which, unlike Sunset Park, has already undergone significant progress and transformation. Rather than looking for obvious signs of growth, such as new construction sites, I focused on something more subtle; areas where development had stalled in otherwise thriving neighborhoods.
Walking along Bedford Avenue, I noticed small but telling differences between each block. One block may feature fully renovated buildings and a strong retail presence, while just a block over, there are much older mixed-use properties with little to no upgrades and sometimes underutilized upper floors. The area, at large, was not underdeveloped by any means, which meant these underutilized properties had significant investment potential. This taught me to look more closely at the “stubborn” points that had not yet been touched by the momentum of progress. Just east of Bedford, for instance, I noticed buildings with outdated, worn exteriors and older retail shops that didn’t match the surrounding atmosphere. In a neighborhood where demand is already so high, these seemingly minor inconsistencies can represent opportunities for strategic repositioning.
This approach shifted how I interpreted behavioral signals. In emerging neighborhoods, I looked for signs of growing confidence, but here the signal was often the opposite: where are owners not acting yet? A lack of upgrades in a high-demand area can indicate hesitation or constraint, both of which have the potential to create investment opportunities. The key in popular areas like Williamsburg is to identify where growth has already been proven but hasn’t fully spread, underscoring that block-by-block analysis should never be a one-size-fits-all approach.
Week 8: AI and Long-term Patterns
For this week, I decided to focus on a broader perspective. My goal was to understand how neighborhoods evolve over time. Instead of only analyzing current conditions, I studied long-term development patterns over roughly 20 years in Canarsie, Brownsville, and Sunset Park. What stood out to me was that development rarely happens all at once; rather, it comes in waves. Usually, it begins with subtle signs like small renovations, increased building permits, or gradual demographic shifts before becoming more visible. To better identify these patterns, I decided to utilize AI to organize and interpret large amounts of data. My prompts were as follows:
“Identify long-term development trends in Brownsville, Brooklyn, using building permits and population data from 2000–2025.” and “Compare early indicators of growth in Sunset Park vs. Canarsie.”
Although AI was very helpful in spotting general trends, it was still important to fact-check its findings. So, I cross-referenced the responses with NYC building permit records, zoning maps, and census data. In some cases, I noticed that the AI overgeneralized or missed local nuances. For example, it interpreted increases in permits as signals of growth; however, those permits could be for minor repairs or safety compliance upgrades, not major developments. This reinforces that AI should be used as a tool rather than an absolute authority.
Through this process, I found that Sunset Park showed earlier signs of steady reinvestment, while Canarsie and Brownsville reflected slower, more gradual change. However, even in these areas, subtle indicators—such as rising permit activity (when backed by fact-checking)—suggest potential for long-term growth.
This week showed me that real estate is not just about where development is happening now, but where it’s going next. Combining block-level insights with long-term data and AI analysis provides a clearer, more forward-looking investment strategy.
Week 9: Going from Patterns to Predictions
This week, I decided to move beyond simply identifying development patterns and instead focus on long-term trends in order to make predictions. While it was interesting to identify major development patterns last week, I wanted to apply the same tools with the goal of predicting where the next wave of development may take place. As such, I adjusted my AI prompts to reflect this new objective:
“Which areas in Brownsville are most likely to see new development in the next 5–10 years based on previous trends?”
“Which areas in East New York have experienced significant growth in the last decade, but where development now seems to be slowing down or coming to an end?”
AI was useful in identifying neighborhoods where development was still ongoing but had already gone through several stages. To ensure that the information I gathered from my searches was accurate, I compared my findings with zoning laws, permits, and the firsthand observations I made during my walks through different parts of the city. I noticed that the AI sometimes highlighted areas that initially looked promising but, upon closer inspection, did not accurately reflect real-life conditions. For instance, a block could appear to show low development, but in reality that was due to zoning restrictions or a lack of demand.
This week taught me that predicting the future requires not only data, but also context and a significant amount of human discernment. Overall, I am beginning to understand the direction my project is moving toward. It is becoming more than simply looking at the past and analyzing it; it is about utilizing a variety of sources in order to make more accurate real-estate predictions.