Airgap Scores Sensitive Tabular Data Fully Offline on a Local GPU

Airgap is an open-source web dashboard that turns CSV or Excel files into risk scores without sending any data to the cloud. It uses TabPFN, a tabular foundation model, to score each row on the user's own GPU, and a local language model then writes a one-page summary. The tool targets small teams such as clinical labs, manufacturing plants, and credit unions that hold sensitive records they cannot upload to external services. A demo on a 30-day hospital readmission dataset reached a cross-validated AUC of 0.67, which the author notes is realistic for a noisy outcome, and it took about 269 seconds on a laptop RTX 4090. The author stresses that the scores are decision support rather than decisions, and that the business model of selling per-site licenses is an untested hypothesis.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)
Log in to join the discussion and vote.
Log in