Loops HouseCachet

NEAR HACKATHON
Team
1 member- BEOwner
Bekir Erdem
Overview
People leave the hospital with papers they only half understand, and the mistakes happen at home: across studies of adults after discharge, the median share of patients with a medication error was 53% (Alqenae et al., 2020). Chatbots that rewrite such papers drop things, and the paper ends up on someone's server.
Cachet turns a photo, scan or PDF of a discharge letter, prescription or lab report, in any language, into a home plan in the reader's language: medicines as printed, check-ups, warnings, and questions for the doctor. The model is never asked to rewrite. Two vision models copy the page independently, code compares the copies and checks every entry against the page, and the person confirms; every plan entry points at its line on the paper. It explains only: no diagnosis, no treatment advice.
NEAR's role: the page is encrypted in the browser to a key that exists only inside a NEAR AI Cloud enclave, after the browser itself verifies the Intel TDX and NVIDIA attestation; our relay only sees ciphertext. Each answer is signed inside the enclave, and a contract on NEAR testnet verifies that signature and records the reading (hashes only). No account, nothing stored. Live demo: https://cachet.l3ekirerdem.workers.dev