House

House helps Wake County buyers understand how much of a property’s assessed value comes from the land versus the building.

Comment

Inspiration

As college students getting closer to graduation, we keep hearing the same thing: housing prices are going up, rent is expensive, and owning a home feels further and further away.

That made us think about housing differently. When someone looks at a property online, they normally see one number: the price. But a property is really made up of two things: the land and the structure sitting on it.

In a growing area like Wake County, there are places where the land itself makes up a surprisingly large portion of a property's assessed value. For a first-time buyer, that matters. You may think you are paying for a house, when a large part of what you are actually paying for is the location and the land underneath it.

That became the idea behind House or Lot.

Our goal was to use geospatial data to make housing data easier to understand and answer a simple question:

Are you buying a house to live in, or a valuable lot that happens to have a house sitting on it?

More importantly, we wanted to show where these patterns are happening, because housing affordability, redevelopment, and land pressure are not evenly distributed across a city.

What it does

House or Lot is an interactive geospatial application that analyzes properties across Wake County using public parcel and Census data.

For every property, we separate the county-assessed land value from the building value and calculate:

$$ \text{Land Share} = \frac{\text{Land Value}}{\text{Land Value} + \text{Building Value}} $$

We then turn that number into a simple, understandable signal:

  • Under 40% land share — House: More of the assessed value is coming from the structure.
  • 40% or greater — Lot: A large portion of the assessed value is coming from the land itself.
  • 40% or greater + built in 1975 or earlier — Teardown Watch: An older structure on relatively valuable land may be worth investigating for possible redevelopment pressure.

These categories are screening tools, not predictions. We are not claiming that a house will be demolished or that an assessed value is the same as a current sale price. Instead, we surface properties and locations where the relationship between land and structure is worth looking at more closely.

The map is what makes the project useful. Instead of seeing isolated tax records, users can visually identify clusters of high-land-share properties and compare them with the surrounding housing environment.

Why this is a geospatial problem

The number by itself is only part of the story. Location changes what the number means.

If multiple older homes with high land-share values begin appearing in the same part of Wake County, that can reveal a geographic pattern that would be difficult to notice by looking at individual property records.

That directly connects to the Center for Geospatial Analytics challenge. We are using location-based data to better understand a pressing societal issue — housing affordability and the changing value of residential land — and then using a map to help determine where people should investigate further.

For a homebuyer, that action could mean researching a property more carefully before making an offer. For planners, researchers, or community members, it could mean identifying areas where older housing stock may be experiencing greater redevelopment pressure.

Instead of simply asking "How expensive is this house?", House or Lot asks:

Where is the land becoming more valuable than the home sitting on it, and what could that mean for the people trying to live there?

How we built it

The core of the project is built around public data rather than generated or imputed housing values.

We use Wake County Property and Parcel data to retrieve information such as land value, building value, year built, heated area, recent sale information, and parcel location.

We combine that with U.S. Census TIGER/Line boundaries and American Community Survey data, including housing tenure, home-value, and rent-burden information. This gives us neighborhood-level context while keeping Census statistics at the geographic level where they were actually published.

On the geospatial side, we process and join location data in Python using tools such as GeoPandas and Shapely. The results are delivered to an interactive web map where users can explore parcels, filter results, select properties, and understand why a location was classified the way it was.

We also built additional technology around the core geospatial analysis:

  • Google Gemini API helps turn raw parcel and geographic information into understandable explanations instead of forcing users to interpret rows of government data themselves.
  • ElevenLabs gives those explanations a natural voice, allowing property information and insights to be narrated instead of only displayed visually.
  • Tiger Data / PostgreSQL provides a structured foundation for storing and querying property and analysis data efficiently as the project grows.
  • Solana gives us a way to experiment with creating verifiable records of selected property analyses or exported results without changing the underlying public source data.

One design decision was especially important to us: AI does not decide whether something is a House, Lot, or Teardown Watch. Those classifications come from transparent calculations using public data. AI is used to help explain the results, not invent them.

Challenges we ran into

One of our biggest challenges was that government datasets are powerful, but they are not designed like consumer applications. Parcel records, Census geography, property values, and map coordinates all have to be cleaned and connected before they become useful to a person looking at a map.

We also had to be careful about what the data actually tells us. A county assessment is not necessarily the same thing as today's listing price. A high land share does not prove that somebody will demolish a house. Census tract data describes an area, not an individual property.

It would have been easy to make the project sound more impressive by turning those signals into predictions. Instead, we focused on making the limitations visible and keeping the calculations explainable.

Another challenge was performance. Geospatial datasets can become large very quickly, so we had to think about how much data to process, what information should be sent to the browser, and how to keep an interactive map responsive.

Finally, integrating multiple APIs and services during a hackathon forced us to constantly decide what actually improved the product versus what was simply another feature.

Accomplishments that we're proud of

  • Every core value traces back to a public dataset, and the main classification can be understood with one formula.
  • AI explains the results but never decides them. House, Lot, and Teardown Watch come from transparent calculations, not a black-box score.
  • We resisted turning signals into predictions. We kept the limitations visible and the project honest about what the data can and cannot say.
  • We turned raw government records into an interactive map where users can see clusters of high-land-share properties instead of isolated tax records.
  • We integrated multiple services (Gemini, ElevenLabs, Tiger Data / PostgreSQL, and Solana) around a geospatial core, while staying selective about what actually improved the product.

What we learned

The biggest thing we learned is that a simple number becomes much more powerful when you can see where it is happening.

Looking at one property's land value does not tell much of a story. Seeing groups of older homes where land represents a growing share of assessed property value creates a question worth investigating.

We also learned how important transparency is when working with housing data. A complicated machine-learning score might look impressive, but for a decision as personal as buying a home, being able to explain exactly where a number came from is more useful.

Every core value in House or Lot traces back to a public dataset, and the main classification can be understood with one formula. That was intentional.

What's next for House or Lot

The next step is expanding House or Lot beyond a single snapshot of Wake County.

We would like to add zoning, permit activity, flood and environmental constraints, and historical assessment data so users could see not only where land pressure exists, but how it changes over time.

We also want to expand beyond Wake County and make the same approach work anywhere reliable parcel and Census data are available.

Ultimately, House or Lot is not trying to tell someone whether they should buy a property. It is trying to give them something most housing websites do not:

A clearer picture of what they are actually paying for, where that pattern is happening, and where they should look deeper before making one of the biggest financial decisions of their life.

Built With

Share this project:

Updates

Submission history