parshotam94/cv_assurance ? reverse-engineered prompt

Build me a fully offline computer vision assurance app that runs locally with a FastAPI backend and a simple browser dashboard.

I want to upload datasets and models, inspect them for problems like duplicates, label noise, out of distribution samples, trigger patches, model tampering, and drift, then see a clear risk summary and findings page. It should also support inference signing and verification so every result can be tied to the input, the model, and a tamper evident audit trail. Please add local storage with SQLite, generate cryptographic keys on first run, and keep everything air gapped with no external calls or CDN assets.

The app should include pages for datasets, models, inference, audit, and reports, plus a main overview dashboard. I also want offline report export to PDF, JSON, and CSV. If anything needs current library details, look up the latest docs online first, then keep the whole thing runnable from the repo with a simple local start command.