Sample Collection Desk

Six diagnostic-centre inboxes receive appointment emails from insurance TPAs in a dozen different formats. An LLM reads each one, extracts the patient, tests, address and time slot, and files it as a case — replacing the manual re-typing of every booking out of email and into a sheet.

Extraction maps a dozen TPA formats onto one schema, including the awkward ones — where a provider identifies the case by an employee ID instead of an application number, or carries the reference number only in the subject line and never in the body.

Coordinators allot cases to field technicians and hand off the day’s visits over WhatsApp in one tap. Technicians confirm each appointment from their phone, and the desk sees the status change on the same record instantly.

One case record is shared by every role, so a technician marking a visit confirmed — or the MER as sent — is visible to the coordinator on the same row. No reconciliation, no duplicate records, no phone calls to ask who is doing what.

Deletion is a soft delete into a recycle bin recording who and when. Only an administrator can destroy a case, and only from the bin.

Guards against silent data corruption: confirmation emails no longer reopen completed work, duplicates are matched and merged rather than re-created, and insurer names that differ by a single character — distinct companies under near-identical branding — are canonicalised separately so they can never collapse into one.

Per-technician dashboards show assigned, completed and outstanding work daily, weekly and monthly, colour-split so a bottleneck is obvious at a glance.

A nightly job rolls cases older than 90 days into aggregate counts and deletes the patient records, so reporting survives but personal data does not.

Deliberately dependency-light — the Excel export is a hand-written XLSX writer rather than a library, because the only spreadsheet package on npm carries unfixed high-severity advisories and this application holds patient data.