Skip to content
Waseem Labs

Medical Fax Platform

Status: In developmentBuilt for NJOSM Clinic · New JerseyWith Omar Khatib

Software that reads a busy practice’s incoming faxes, works out which patient each one is about and what kind of document it is, and files it in the right place.

Illustrative · How it works: Medical Fax Platform. An illustrative diagram, not a product screenshot.

Today, without it:

  • fax arrives
  • opened by hand
  • “who is this about?”
  • filed or forwarded by hand

With the platform:

  1. Fax arrives (Queued the moment it lands)
  2. Every page read (Text taken from each page)
  3. Which patient? What is it? (by AI; One of 35 document types)
  4. Patient’s folder (Filed and indexed)
  5. If can’t tell: Unassigned (Nothing is lost)
  6. Staff (From the same system)
  7. Send or forward (By fax, email, or text)

The hard part: A fax the system can’t place is never lost.

The problem

A busy orthopaedic practice still runs on fax. Referrals, authorisations, reports, and records arrive all day as image-only PDFs, and someone has to open each one, work out who it is about and what it is, then file or forward it by hand.

What it does

  • Every incoming fax read automatically, page by page
  • The patient identified, and the fax filed in their folder
  • Each fax sorted into one of 35 document types, from referrals to prior authorisations
  • Faxes that need to go elsewhere flagged for forwarding
  • Replies sent by fax, email, or text from one place

A pipeline that reads like the process

When a fax arrives, the webhook is acknowledged immediately and the work is handed to a Cloud Tasks queue, so slow OCR never holds up the fax provider. Each fax then moves through single-purpose stages — fetch, read, extract, classify, file, index — and a fax that was already processed is detected and skipped, so a repeated notification cannot create a duplicate record.

AI that returns structured answers

A fine-tuned Llama 3 70B reads the extracted text and answers in a fixed, machine-readable format: the patient, their date of birth, and whether the fax should be forwarded. It then classifies the fax against 35 document types built around the practice’s real workflow, from referrals to prior authorisations. A fax the model cannot place is filed as Unassigned, so nothing is lost.

Integrations

  • Fax.Plus for receiving, sending, and forwarding faxes
  • Text messages through GoTo Connect, with OAuth tokens refreshed and rotated automatically
  • Email through the Gmail API, sent from the staff member’s own mailbox
  • Google Chat alerts when a new fax arrives

The diagram on this page is an illustrative system flow, not a screenshot of the product.

Have a problem like this one?

Tell me how the work happens today. We’ll work out what’s worth building.