Zonar

Inspiration

New businesses and families needs easier ways of finding proper places to start their beginnings. Small businesses typically spend weeks to months scouting locations, piecing together demographics, foot traffic, and competition manually. Enterprise site selection tools are expensive and typically made for larger chains. Using agentic LLMs with free census data can solve this: a platform that allows anyone to find locations that are maximized to their profitability based on the client's business in minutes for far less than is being offered currently, allowing the client to truly get started instead of waiting for the right moment.

What it does

Zonar takes a business idea and the city they want to be in, then analyzes hundreds of zones using real census data to find the best areas to rent/buy in, with proper scoring by foot traffic, demographic fit, competition gap, rent efficiency, and walkability. Using multi-agent AI, we return a ranked heatmap of the best zones to open, plus AI insights for each top zone.

How we built it

Zonar uses a multi‑agent pipeline plus cached citywide data on a 5-layer architecture:

Query Parser (Gemini): Turns the business idea into structured parameters and search terms (ex: "Bubble Tea Shop in Providence" into business_type, price_point, peak_hours, etc).
Market Scanner (data): Loads cached citywide Census + OSM from S3 (no live API calls). Provides the base data for scoring every block group
Financial Modeler (heuristics): Estimates revenue, rent, and breakeven per location/zone.
Location Scorer (heuristics): Computes PlotScore from subscores and user‑defined weights.
Insight Explainer (Gemini): Generates AI narrative for top zones (prefetch) and on click.

On‑click Web Researcher (Gemini + Search grounding): Pulls market context (rent ranges, regulations, openings/closures) only when a zone is selected.

Frontend: Next.js + Mapbox heatmap + slider‑based reweighting Backend: FastAPI on AWS Lambda + DynamoDB + S3 caching

## Challenges we ran into

Figuring out proper caching bulk for Census/OSM data without hitting API limits
Making AI feel “instant” on a heatmap without slowing first paint

Accomplishments that we’re proud of

Heatmap scoring across ~200 block groups in ~2 seconds
Click‑based AI insights and market intel
Fully serverless backend + cached citywide data pipeline
User‑adjustable weighting that re‑ranks zones in real time

What we learned

Caching citywide data is absolutely needed in order to be able to run fast queries
Zone‑based ranking is more useful than listing existing businesses for new owners