-
Home page
-
311 query example
NYC Urban Intelligence
Inspiration
NYC provides a high volume of open source data, so much so that search, discovery, and analytics can be difficult for individuals. We endeavored to create an agentic natural language framework for people to explore and obtain insights from NYPL and NYC Open Data.
What it does
NYC Urban Intelligence is an agentic system comprised of an orchestrator agent, expert agents, and a dyanamic agent driven UI. Given a user prompt or request, the agentic system dynamically fetches and visualizes the information.
How we built it
We built a full-stack web application using Python for the backend and Javascript and Vite for the frontend. Our agentic structure follows what is generally an A2A structure, with an orchestrator and expert agents, each with a set of tools. The frontend interface leverages the A2UI protocol to enable dynamic generation of a bespoke tileset (e.g., text, graphs, maps) that is presented to the user. The full stack application is hosted and deployed via GCP, using serverless Cloud Run. The use of standard protocols and a well constructed agentic structure (agents, tools, and UI) enable extensibility and development beyond this hackathon.
Challenges we ran into
Setting up deployment infrastructure via GCP was a challenge.
Accomplishments that we're proud of
We are proud of the extensible structure of the code and the agentic structure as it adopts modern structures and schemas.
What we learned
We learned that coordination across such dense data is difficult, and it is important to tightly scope our project and start small.
What's next for NYC Urban Intelligence
We are working to integrate Fish Audio to enable audio-to-text input and text-to-audio responses from the agent.
Log in or sign up for Devpost to join the conversation.