Habitat Watch

Logo

Inspiration

The idea for Habitat Watch came from problems we experienced in our own residence hall at Georgia Tech, Bud and Val Peterson.

There was a rat infestation on another floor of the building that eventually spread to our floor. The problem was that we had no idea residents on another floor were already dealing with it until it had reached us. That made us realize how disconnected people in the same building can be when it comes to housing issues.

We also noticed how slowly some maintenance problems can get resolved. Our showers went without hot water for around three days before the issue was fixed, and some hand dryers in the building have remained broken for weeks.

These experiences made us ask a simple question: what if residents could actually see the problems other people in their building were experiencing, document their own issues properly, and recognize when something was becoming a larger shared problem?

That became the idea behind Habitat Watch.

What it does

Habitat Watch is an AI-powered housing platform that helps residents document housing problems, understand what they are seeing, and connect with other people experiencing similar issues.

A resident can upload or take a photo of a problem such as mold-like growth, water damage, pests, electrical hazards, structural damage, or security issues.

The image is analyzed using a multi-stage AI pipeline. Habitat Watch identifies the visible issue, highlights the affected area, and creates structured evidence around the report.

Residents can then see reports from others in the same building or area through a Community Timeline, where issues are organized into categories such as water and moisture, pests, electrical, structural, and security.

Habitat Watch can also identify possible shared patterns across multiple reports. It generates an AI summary of what residents are experiencing, shows the reports and evidence behind that summary, and visualizes where issues are occurring while allowing residents to remain anonymous.

The platform also includes a legal-information layer that connects documented conditions to verified local housing resources and helps residents understand what additional evidence they may need or how to structure a repair request.

The goal is to turn isolated housing complaints into organized, shared information that residents can actually use.

How we built it

We built Habitat Watch as a responsive web application using Next.js and React, allowing the same platform to work on both desktop and mobile browsers.

We used Cursor Pro extensively throughout the hackathon to help us rapidly build, debug, and iterate on the application.

For the computer vision system, we used Meta Muse Spark 1.3 and SAM 3.1 in a multi-stage pipeline.

First, Muse Spark 1.3 analyzes the full image and determines what it is looking at. This allows it to distinguish normal objects such as lights, vents, or shadows from possible housing issues.

Muse then passes the suspected issue to SAM 3.1, which performs the segmentation step and highlights the relevant region of the image.

Finally, we send that detected region back through Muse Spark 1.3 for a second verification step. This helps confirm that the highlighted area is actually a housing issue rather than a normal object or visual artifact.

We also use Grok for the legal-information side of Habitat Watch. Grok works with verified housing-code and local housing-resource information to help connect a documented condition to potentially relevant rules, identify useful additional evidence, and assist with structured repair documentation.

Challenges we ran into

The biggest challenge we faced was making the image analysis accurate enough to actually be useful.

Early in development, the segmentation could sometimes highlight too little, highlight too much, or identify the wrong part of an image. In one case, an ordinary light could be mistaken for discoloration.

We realized that segmentation by itself was not enough. The system needed to understand the entire scene before deciding what should be highlighted.

That led us to redesign the computer vision pipeline into three stages:

Muse Spark 1.3 → SAM 3.1 → Muse Spark 1.3

Muse first understands the scene, SAM precisely segments the suspected condition, and Muse checks the result again.

This significantly improved the system because each model now has a specific role instead of relying on one model to do everything.

Another challenge was making the community part of Habitat Watch genuinely useful. We did not want it to become just a list of words like "mold," "leak," or "pest." We wanted residents to actually understand what other people were experiencing.

That led us to build detailed reports, the Community Timeline, anonymous submissions, AI-generated shared patterns, evidence-backed summaries, and visualizations showing where problems are occurring throughout a building.

Accomplishments that we're proud of

We are most proud that Habitat Watch solves a problem that we have personally experienced while also bringing people together.

What started as an idea about detecting housing problems became something much larger. Habitat Watch allows residents who may never normally speak to each other to understand that they are experiencing the same problem.

We are also proud that the AI is not just included for the sake of having AI. Each part of the system has a clear purpose: understanding the image, segmenting the issue, verifying the result, finding patterns between resident reports, and helping people better understand their housing situation.

Most importantly, we built something that we would genuinely want to use in our own residence hall.

What we learned

One of the biggest things we learned was how little visibility residents can have into problems happening only a few floors away from them.

Before working on Habitat Watch, we thought about housing problems mostly as individual maintenance issues. During the project, we started thinking about them as building-level problems that can affect many people without those residents ever realizing they are connected.

The rat infestation in our own building made this especially clear. If residents had known what was happening on another floor earlier, they could have been more aware of the problem before it spread.

We also learned that good documentation matters. A single photo may not tell the full story, but multiple timestamped reports from different residents can reveal something much more meaningful.

That changed the way we thought about the project. Habitat Watch became less about simply detecting damage and more about giving residents awareness of what is happening around them.

What's next for Habitat Watch

We want to continue developing Habitat Watch beyond the hackathon and explore turning it into a real product.

Our next goal is to pursue the idea through Georgia Tech's CREATE-X program and explore how Habitat Watch could grow into a startup.

We would like to expand support to more buildings and cities, add more local housing-code information, improve the image-analysis system across more categories of housing problems, and make the community and shared-evidence features even stronger.

Long term, we want Habitat Watch to become a platform that helps residents go from noticing a problem to documenting it, understanding it, discovering who else is affected, and taking informed action together.