NYC Storefront Database Updater

AI-assisted database maintenance tool, allowing a non-profit to maintain an accurate, continuously updated database of their neighborhood's storefronts.

Category
Other Projects
Visit Demo Site
NYC Storefront Database Updater hero image

Case Study

Challenge

The East Village Community Coalition maintains a 1,800+ database of neighborhood storefronts, which serves as a crucial tool for neighborhood surveys and advocacy, tracking trends such as different business types' relative longevity, emerging corridors of inactivity, and problematic landlords. Keeping it up-to-date meant manually tracking openings and closings on a spreadsheet across a dense neighborhood with frequent turnover. Google Maps tells you what's at an address today, not what was there before, when it changed, or the source of the information.

Solution

A system parses six sources for signals — local news, Instagram, business websites, Google Places, and the NYC DOHMH inspections feed — and provides suggestions in a queue to review. LLMs act as a research assistant, extracting business event(s) from each story (open / close / what it's replacing / opening date), matches it to the right physical storefront, and suggests values for key fields like business category, based on available signals. Crucially, a human reviewer confirms or edits every change before it's written to the database.

Results

For three years, we have maintained a neighborhood-scale database without a neighborhood-scale team, transforming what used to be drawn-out manual work (find the article, parse the event, geocode the address, identify the previous tenant, propose a category) into a clean queue of pre-drafted suggestions, allowing one person to accurately and consistently maintain the entire database.

Implementation

Built solo, I created a worker script running on Cloudflare Workers, with a React frontend for reviewing the suggestions. Five types of LLM calls are wrapped in deterministic scaffolding: strict tool schemas, closed enums, dynamically-generated few-shot exemplars, and cited sources that are checked, to reduce hallucinations. The key step to ensuring accurate location data is a combination of a street-normalization utility script and NYC GeoSearch/Pelias API, which transforms casual references in the source data into real addresses that city government understands.

Impact

Empowers the East Village Community Coalition to participate in community planning and advocacy by providing them with a continuously current, fully sourced business dataset, making storefront surveys possible on demand, rather than something that's done partially every five years.

Tech Used

Cloudflare WorkersCloudflare D1AirtableNYC GeoSearch (Pelias)Google PlacesOpenAIGoogle GeminiInstagram ScraperReact
Visit Demo Site