FLORIDAPLATFORM / HOW IT WORKS
Useful software. Visible evidence.
We build software around recurring Florida public-record research. This site demonstrates the records, decisions and review steps that can become part of a customer’s workflow.
Follow the records.
The restaurant examples compare two fictitious-name filings with food-license records. We checked the original Sunbiz daily-file lines and DBPR rows, retain file checksums, and show the publication and retrieval dates. Both examples were rechecked September 20, 2026. The food-license extract was published September 14; the Sunbiz filings are dated September 18.
The comparisons are selected, curated demonstrations. The featured businesses are not customers. A shared legal name creates a candidate connection; it does not prove current operation, an ownership transfer, or a completed licensing change. We keep those questions visible.
Ask a question. Inspect the interpretation.
The live table sends your question and a bounded capability description to hosted TypeSafe Jev through OpenRouter. Jev chooses supported filter conditions; ordinary code applies them to a snapshot of 15,681 permanent food-license records in Miami-Dade, Broward and Palm Beach. Filter chips show what was applied. Reused model responses are marked as cached.
Supported questions include county, names containing pizza or coffee, at least 100 seats, missing recorded inspection dates, and different licensee/business names. Exact-filter mode also supports specific name, seat and date filters locally. This snapshot cannot establish opening dates, closures, permit history or purchasing intent.
Where our trained model fits.
Florida Project Model v5 is our separately trained small Qwen3-0.6B adapter for narrow permit-reference and filed-role decisions. It is evaluated locally and is not deployed on this public site. It is not TypeSafe’s proprietary Jev model. A customer workflow can combine exact checks, model suggestions and human review; training is justified by measured failures, not by default.
Start with a measured pilot.
We begin with one repeated task, named sources and jurisdictions, and agreed examples. A two-week pilot starts from $2,000. We compare staff time including review, missing required records and incorrect matches. Production setup and ongoing hosting, refreshes and support are quoted separately. There is no automatic subscription.
Current examples are source-dated snapshots, not continuously refreshed statewide data. Each production workflow needs an explicit coverage and refresh agreement. We do not promise government approval, current ownership conclusions or legal advice.
Bring your existing tools.
Delivery can be a private research interface or an integration with your team’s case tracker. We agree on inputs, outputs, access and review requirements before using client data. Private documents and client data are not needed for the introductory conversation.