8
Products shipped
AI for complex, legacy systems.
The hardest data in the world is public — and almost unusable. We rebuild it into software anyone can use.
NYSGPT builds domain-specific AI applications for the systems ordinary software won't touch — the impenetrable government databases, statutory regimes, and industrial records that ordinary tools leave untouched. We turn data nobody could read into software anyone can use.
The most consequential systems run on the worst software. Legislative records, customs rulings, energy filings, licensing registries, nuclear references — the data is public, authoritative, and locked behind portals built two decades ago for nobody in particular.
NYSGPT exists to close that gap. We don't bolt a chatbot onto a problem. We ingest the primary record, ground every answer in a verifiable source, and rebuild the interface from the data up — so a stressed parent, an importer, a researcher, or a regulator gets a real answer instead of a 40-page PDF.
350k+
Records navigable
5
Live in production
1
Founder · full-stack
We don't bolt a chatbot onto a problem. We build from the record up.
01Grounded, not generated. Every claim traces to a primary source. Fluency without traceability is a liability.
02The data is the product. We rebuild the interface from the record up — search, structure, and AI as one surface.
03Ship to production. Live products with real users, not decks. Five of eight are running right now.
The most consequential systems run on the worst software. Legislative records, customs rulings, energy filings, licensing registries, nuclear references — the data is public, authoritative, and locked behind portals built two decades ago for nobody in particular.
NYSGPT closes that gap. We ingest the primary record, ground every answer in a verifiable source, and rebuild the interface from the data up — so the person who actually needs the answer gets one, not a 40-page PDF.
Each product below is the same architecture, drawn for a different domain: ingest the hard public record, make the AI accountable to it, and ship the result to people with real problems.
Register
№
Product
What it does
At scale
Status
01
44B
A public registry, research library, and incident record for AI — the compliance portal for NY's Article 44-B. "SERFF, for AI."
32k papers+2,561 incidents
LIVE
02
TariffsGPT
A self-serve refund engine for importers who paid IEEPA tariffs the Supreme Court invalidated — a real number before any phone call.
220k+ rulingslive tariff schedule
LIVE
03
ChildcareGPT
NY childcare intelligence — every OCFS-regulated facility, county cost data, and an AI-assisted subsidy application form.
16,800 facilitiesOCFS-6025 form
LIVE
04
SolarGPT
Clean-energy intelligence — solar potential, grid economics, and emissions for every U.S. county, state, and grid region.
3,000+ countiesEIA data pipelines
LIVE
05
PolicyGPT
NY legislative & financial intelligence — bills, budgets, lobbying, contracts, and members, with an embedded advocacy layer.
Full NYS recordsemantic search
SHIPPED
06
NSRgpt
Semantic search and AI chat over the Nuclear Science References database — vector embeddings and a full RAG pipeline.
88,000+ recordspgvector RAG
SHIPPED
07
InsuranceGPT
An auto-insurance quote & lead engine — rate estimation grounded in filed carrier data, Stripe checkout, multi-role console.
Grounded ratesStripe + console
SHIPPED
08
SportsGPT
High-school sports intelligence for Section XI — team and player stats, standings, choropleth maps, AI-generated highlights.
Section XIAI highlights
SHIPPED
Domains we've already gone deep on
Government & legislative
Trade & customs law
Energy & grid economics
Social services
Insurance & regulation
Scientific reference
AI governance
Public records
How we build
01 — INGEST
Tame the primary record
We crawl, dedupe, and structure the authoritative source — even when there's no bulk export, no API, or a pagination wall — into clean, queryable Postgres. The hard part nobody else does.
02 — GROUND
Make the AI accountable
Retrieval over the real corpus, structured output, and a visible chain back to the source. The model reasons over primary data, and every answer can be checked.
03 — SHIP
Rebuild the interface
Search, structured exploration, maps, and AI on one surface — designed for the human who actually needs the answer, and pushed to production with real users.
Letters to the editor
Have a system that's too complex for ordinary software?
That's exactly what we build for. Book a working session and we'll pressure-test what an AI-first rebuild looks like for your data — or send a short brief and we'll come back with a concrete read.