Where the data comes from, how often it refreshes, and what we do not know.
DSNY's Executive Director of Resource Recovery asked how the agency could improve its existing How To Get Rid Of tool — in practice a 63-row CSV powering a simple keyword redirect on nyc.gov — into a user-friendly, mobile-first AI digital tool that helps New Yorkers recycle right and supports the City's 2026 Solid Waste Management Plan diversion goals. The ask was for a tool that could integrate with the City's IT infrastructure and be widely available across platforms, built with a partner experienced in app-building and NYC government work. DSNY offered the current tool's underlying materials data and IT/security guidance in return.
Sponsor: Kate Kitchener, Executive Director, Resource Recovery · NYC DSNY
Get Rid Of It Right is a grounded AI answer engine: a visitor asks in plain language how to get rid of or donate any item, and it returns a cited answer sourced only from DSNY's own published Get Rid of pages — never improvised — paired with a real map or list of nearby donation, special-waste, or food-scrap drop-off sites when applicable. It declines to answer rather than guess for unsafe or out-of-scope items, and surfaces reuse and donation options ahead of disposal for donatable categories like clothing, furniture, electronics, and appliances. It's a single, anonymous public flow with no login or staff console.
Click any freshness pill on any page to jump directly to the relevant source row here.
We scrape the 19 core DSNY category pages plus the peripheral pages referenced by DSNY's own search index. Every scrape produces a new content-hashed corpus version so we know when the underlying guidance changes.
The DSNY corpus and item index are English only. Multilingual support is on the product roadmap.
Answers ground in the 33 pages DSNY publishes publicly. Item-level guidance that lives inside DSNY's internal materials database is out of scope.
We render hours verbatim from the city dataset. We do not attempt to compute whether a site is open at the current time.
The product is read-only. For enforcement, missed collection, or emergency guidance, contact 311 directly.
The SWMP 2026 rulemaking PDF is policy framing rather than per-item guidance. It is not part of the RAG corpus.
About 57 percent of donation-site rows have a normalized category tag. The rest fall back to the raw string DSNY publishes, which the UI shows verbatim.
Retrieval uses voyage-3-lite embeddings. The similarity floor is calibrated to keep off-topic queries out of the answer path while preserving on-topic retrievals.
Live donation sites accepting clothing, furniture, electronics, and household items, published as an NYC Open Data set. We only render rows with valid latitude and longitude.
City-run Special Waste drop-off sites. Five sites citywide, one per borough. Sites accept harmful household products, batteries, and other flagged waste by schedule.
Live neighborhood food-scrap drop-off locations, published as an NYC Open Data set. We only render rows with valid latitude and longitude.
The retrieval index used by the answer service. Each corpus version is chunked and embedded so grounded answers can cite the exact DSNY passage they draw from.
Answers are generated by Anthropic Claude using the DSNY passages retrieved with Voyage AI embeddings. Only the retrieved DSNY passages are shared with Claude at request time. No user identity, precise location, or account information is passed.
Answer generation is optional. Category browse and near-me drop-off lookup continue to work when the answer model is unavailable. NYC GeoSearch autocomplete is a live lookup with no persistence and requires no key.