Cascade

Main page

Inspiration

Policy debates are usually loud but hard to follow. Most people can hear a headline about a tariff, tax, or reform and still not answer the practical question: what will this actually change for real people? We built Cascade to make those consequences visible in seconds, not hidden in 80-page reports.

What it does

Cascade takes a plain-English policy description and generates an interactive impact analysis across:

countries and regions
trade and market effects
demographic winners and losers
second- and third-order knock-on effects over time

How we built it

Frontend app with React + Vite
Structured analysis pipeline powered by Claude via the Anthropic Messages API
A strict JSON contract (ImpactSpec) between AI reasoning and UI rendering
Visual layers for map, routes, flow graph, demographics, and timeline using Plotly + D3 + TopoJSON
Fast iteration and UI polish with Tailwind CSS

At a high level, we treat policy shock as a propagation system over regions and causal links:

$$ I_{\text{total}}=\sum_{r} w_r I_r+\sum_{e}\alpha_e\Delta_e $$

This helped us combine direct impacts and knock-on effects in one coherent view.

Challenges we faced

forcing consistently valid structured output from open-ended prompts
balancing speed and depth under hackathon time pressure
representing uncertainty clearly without making results feel vague
keeping visual complexity readable on one screen

What we learned

strict schema design is critical for reliable AI-to-UI products
the biggest value is reasoning quality, not just text generation
policy tools need to communicate caveats and confidence, not only conclusions
fixed visual components are often more robust than fully generated UI code in a hackathon setting