Document text never leaves the machine.
The research build is offline-first by design, so experiments can run on sensitive text without any of it being transmitted. Everything below describes how this project handles data — it is not a compliance claim.
Data-handling principles
Offline-first architecture
The research build runs entirely on local hardware. No outbound connection is required for inference, and no document content is transmitted or logged remotely.
Zero data egress
Document text stays on the machine that processes it. There is no cloud service it reports to, and experiments are routinely run on a network-isolated machine.
Public data only
The research corpus is drawn from publicly published, FDA-approved label documents. No confidential, unpublished, or client-supplied material is used.
Logged, reviewable findings
Every scan and every finding is recorded with a timestamp so that results can be re-examined, audited, and replicated rather than taken on assertion.
Not a validated system
This is research software. It has not undergone GxP validation, it is not qualified for use as a regulatory control, and its output must not be relied on as a quality or compliance decision.
Where the data lives
In the research build: on local hardware, and nowhere else. This public site does not accept document text for processing.