CrystalBall for Cities

Inspiration

Cities often respond to technology too late. We wanted to build a system that could identify weak early signals of technologies before they become mainstream. Instead of asking “what is popular now,” we focused on “what may become important later, even if it is still overlooked today.”

What it does

CrystalBall for Cities is an early-warning system for emerging urban technologies. It tracks candidate topics such as Digital Twin for Infrastructure, AI-based Water Leak Detection, V2G, Drone-based Infrastructure Inspection, and Urban Flood Forecasting, and ranks them based on weak signals from available data sources.

How we built it

We first defined a set of candidate technology topics and created keyword dictionaries for them. Then we used patent data as our main signal source, because it is structured and works well for historical trend analysis. We built a local pipeline using PatentsView / USPTO download tables, joined patent metadata with abstracts, and filtered results using topic-specific matching rules.

We also started exploring how news data and company websites could be added, but this part is still limited in the current version because the data is harder to collect and less complete than we expected.

Finally, we organized the available signals into an interpretable scoring framework based on growth, persistence, cross-source consistency, underattention, and city relevance.

Challenges we ran into

One challenge was that keyword matching is noisy. For example, many patents containing “digital twin” are truly about digital twins, but not all of them are about infrastructure-related digital twins. This meant we had to spend time refining topic boundaries and filtering rules.

Another challenge was data collection. Patent data was relatively structured, but news and website data were much less complete and harder to organize. Because of that, the current project relies more heavily on patent data than we originally expected.

Accomplishments that we're proud of

We are proud that we turned a broad idea into a clearer and more realistic project. We defined a focused problem, selected candidate topics, built a patent-based signal pipeline, and created an interpretable scoring framework. We also clarified how the project can later grow into a stronger multi-source early-warning system.

What we learned

We learned that weak-signal detection is not just about finding trending keywords. It requires careful topic definition, cleaner retrieval, and better data organization. We also learned that explainability matters: it is not enough to output a score, because users need to understand why a technology is being highlighted.

What's next for CrystalBall for Cities

Next, we want to improve the project in three ways: First, we want to make the current patent pipeline more accurate and scalable across multiple topics. Second, we want to add more complete news and website signals. Third, we want to introduce historical benchmark topics and negative examples so that we can backtest and better calibrate the scoring framework.