Atlis started with a simple observation: transportation systems often fail without warning. You might be on a train that suddenly stops because of an issue no one detected early, or driving over a pothole that has likely been there for weeks. In both cases, the problem is experienced, but nothing is reported or captured in real time. Living in New Jersey, where many people rely heavily on both roads and rail, made this issue even more obvious. We realized that infrastructure problems are everywhere, but the systems used to monitor them are outdated and mostly reactive.

This led us to ask a key question: what if every car and train could automatically collect data about infrastructure as it moves? That idea became Atlis. Our goal was to create a system that continuously gathers and analyzes data without requiring any effort from users. Instead of relying on manual reports or scheduled inspections, we wanted a solution that works passively in the background and provides real-time insights.

Through this project, we learned a lot about building systems that work in real-world conditions. We explored computer vision and trained models to detect things like potholes and cracks from video footage. We also learned that technical performance alone is not enough. A solution like this has to fit into existing systems, be easy to use, and provide clear value to governments and users. We also thought carefully about data privacy and fairness, making sure that all data is anonymized and that our system does not create unequal outcomes across different communities.

We built Atlis as a full system with several parts working together. First, we used a Raspberry Pi-based dashcam setup to capture video during travel. Then, we developed an AI model that processes this footage and detects infrastructure issues. Each issue is automatically tagged with location and time data. This information is sent to a central system, where it is displayed on a live map that governments can use to monitor conditions and make decisions.

We were able to build a working prototype in a short amount of time. Within 24 hours, we created a mobile app, set up the hardware system, trained a detection model, and developed a dashboard to display the results. Even though it was a fast build, it demonstrated that our idea could work in practice.

One of the main strengths of Atlis is that it works across both roads and rail. Instead of treating them as separate problems, we designed one system that can handle both. Another important feature is that it runs continuously. The more people use it, the more data is collected, and the better the system becomes. This creates a strong feedback loop where increased usage leads to better insights and greater value.

We also faced several challenges while building Atlis. Detecting infrastructure issues in real-world conditions is difficult because of changes in lighting, camera angles, and road quality. Running AI models on small hardware like a Raspberry Pi required us to optimize for speed and efficiency. We also had to design a system that could handle real-time data without delays. Another challenge was making sure the system required no effort from users, since even small amounts of friction can prevent adoption. Finally, we had to ensure that our approach to data collection was responsible and protected user privacy.

Atlis has the potential to change how infrastructure is managed. Instead of waiting for problems to become severe, governments can detect and address them early. This can lead to safer roads, fewer delays, and better use of resources. Our plan is to start with smaller pilot programs, then expand to larger regions as we prove the system’s effectiveness.

In the end, Atlis is built on a simple idea: every mile traveled can provide useful information. By capturing and using that data, we can help improve the systems people rely on every day.

Built With

Share this project:

Updates

Submission history