Clinical records rescue, before the lights went out.

Medicare requires seven years of records. The company's images, reports and patient records lived in a radiology platform that was being retired, with a hard cutoff for access.

An extraction pipeline that moved every order's images and reports into long-term storage, verified each upload, and kept capturing new records until the last day.

• Hardened against interruption: throughput went from 1,772 to 7,632 orders a day, cutting about 15 days of remaining work to about 3.6 ahead of the deadline
• Every upload verified with a checksum round-trip, and a final reconcile report archived with the records
• A nightly capture that ran three times a day near the cutoff, shrinking the window of unarchived work from about 24 hours to about 10
• Alerts built for silence, so a run that reports success but archived nothing still raises a flag

Before sign-off, the platform changed how image downloads worked, without notice. 96 orders were failing quietly. 93 were recovered, and a spot-check confirmed they were real studies. The final three were canceled orders.

34,784 of 34,787 orders captured (99.991%), 241,000 DICOM files, 835 GB, and 100% of primary clinical reads.

After the old platform was retired, the archive became a read-only, searchable viewer behind company sign-in, with every retrieval logged, for about $12 a month. It now backs 86% of active denial appeals.

With my agentic team, treating "nothing lost" as the requirement: every count reconciled, every gap explained, and the audit trail kept alongside the records. HIPAA is part of the design: patient data minimized, access role-based at the API layer, and a person reviews before anything goes live.