-
-
slide highlighting goals
-
first frontend page
-
chat page with interactive map
Inspiration
Finding apartments as an intern or new grad moving to NYC is a daunting task. I know it. I lived it 4 months ago when I came to the city for an econ consulting role. I would have many pages open on Zillow, late nights compiling spreadsheets, and trying my best to avoid unnecessary transfers but failing. I found ChatGPT not very helpful at suggesting me neighborhoods. Since I had little idea of NYC geography, it did not give me a good sense of distance from work without opening up several other tabs of Google maps. (Real experience from Max and speaking with other new grads in NYC)
This is why we created StationScope, an interactive chatbot robust against AI hallucinations by pulling reliable data from the U.S. Census Bureau and MTA for quantitative data measures. It is a custom AI with weights that guide a user to their dream apartment near a subway stop, making a final recommendation with a dynamic report highlighting points from the chatbot conversation along with information grounded in real-world data.
What it does
It is a fully functional dynamic app. You type in the desired address in the first landing screen, and Google Maps API helps you find a valid address. The app identifies the closest station to your input address as well as maybe additional close stations. Behind the scenes, the app pulls demographic summary stats using a U.S. Census Bureau API for all stations on the same lines as the closest station. We use census tracts partially or fully within a 1/4 mile radius of each subway station and the 2019-2023 5 Year ACS. This information is fed into the bot as reliable quantitative data (less prone to the hallucinations large market AI models can have with numbers). You are prompted to click to on a chat button, after which you can see an interactive map to get a good sense of geographical layout and can chat. The custom AI will iteratively ask you questions and ultimately get you your desired neighborhood/stop and information about that neighborhood/stop in the form of a concise, highly readable three slide report!
How we built it
To develop the app, we had two main workstreams: the frontend + data modeling in R which Max handled, and the agentic bot and map, which Daren handled. The goal is to educate the new mover to NYC about their options and make a structured recommendation, much like a consultant would. Anyone can understand the inputs and outputs to this app, even though the infrastructure is a complex web of Javascript, R, and API links and endpoints.
Challenges we ran into
Coordinating our differing expertises: Daren has more skills in fullstack development, Max has more skills in data science. Smoothly integrating the APIs took some time and a lot of iteration. We slept for 2 hours.
Accomplishments that we're proud of
This is Max's first hackathon!
What we learned
Always plan for more time than it seems. Bugs pop up out of nowhere all the time. R backend plus JSON is an unusual combination but it works with a very helpful R library called "plumber." Commit and push often!!!
What's next for StationScope
Making the AI model even more robust. It turned out pretty well, we think, having only one day (and long night) to work on it, though. We also want to finalize the report output and potentially host the site for the public to use.
Log in or sign up for Devpost to join the conversation.