PROPOSED PROJECT
Patient Referral Tracker
Turn referral documents and scanned forms into a clear appointment request list for clinic staff. Extract contact details and requested appointments, flag duplicates or missing attachments, and show the original document beside each entry for checking.
Every request. With its source attached.
Administrative review only
No appointments booked
Complete drafts
3Staff review still required
MRI appointment request
Required fields present
Physiotherapy appointment request
Required fields present
Consultation request
Required fields present
Needs information
2Keep the missing item visible
Imaging appointment request
Attachment listed but absent
Consultation request
Contact number unreadable
Possible repeat
1Staff decide whether to link
MRI appointment request
Same supplied referral reference
Compare with R-104 ↗Read once.
Check against the original.
The illustrative intake desk receives six synthetic submissions. Three contain the required administrative fields and are ready for staff review. One lacks an attachment, one has an unreadable contact number, and one may repeat an earlier submission. The source panel follows referral R-104 from page 1 to its draft appointment request. No real patient data is used, and no request is treated as an approved booking.
WHAT IS USUALLY DONE
Clinic staff open referral PDFs, read scanned forms, and retype the details into a spreadsheet. Missing pages and repeat submissions can look like new requests. Finding the original evidence for one field means reopening attachments and searching through an email thread.
WHAT SOVRUN CAN DO FOR YOU
Sovrun would build a referral review desk that keeps each proposed appointment request beside its source document. Document models extract the agreed administrative fields, attach a page reference, and flag unclear text, missing attachments, or possible repeats. Staff correct and approve the record before it enters the appointment request list. The first build focuses on intake administration, with clinical assessment and appointment decisions left to the clinic’s existing process.
A field should point
to where it came from.
The original page stays beside the draft. Staff can verify each field without searching another inbox, and incomplete or repeated submissions keep their own place in the queue.
request
- Example patient record
- Sample person 104
- Requested appointment
- MRI appointment request
- Contact number
- 09•• ••• 0104
- Attachments
- Referral form / 1 page
Fictional source. Contact digits masked for display.
Ready to check.
- Reference
- EX-104 Page 1 · source field
- Request
- MRI appointment request Page 1 · highlighted passage
- Contact
- Present in sample source Staff confirmation required
- Attachment
- 1 expected / 1 received
Verify the request and contact details before adding this record to the appointment request list.
Leave the number blank.
An unreadable digit is not an invitation to guess. Preserve the source and use the clinic’s normal follow-up process.
Compare before combining.
The same reference suggests a repeat submission. Both documents remain separate until a staff member confirms the relationship.
A checked request list.
An explanation for every hold.
The first build supports an agreed set of referral forms and administrative fields. It helps staff prepare the appointment request list while preserving the evidence and unresolved work.
Keep the original close.
Nothing disappears.
- R-107
- Missing attachment
- R-108
- Unreadable contact
- R-109
- Possible repeat of R-104
Ready means checked.
in this example
Three complete drafts are waiting for review. No booking action is triggered.
How we’d make it happen.
Define the intake record
Map the actual forms and required attachments. Agree which administrative fields may be extracted, which values require staff confirmation, and where sensitive documents may be processed.
Read the document and keep the source
Extract draft fields with page references. Run completeness and duplicate checks, preserve the original submission, and route uncertain records to a visible review queue.
Approve the request list
Staff compare each draft with the original, resolve missing information through their existing process, and approve the records that can enter the appointment request list. Log the corrections and review decisions.
Where AI helps
We compare document-reading and language models on the clinic’s permitted sample forms, including scans, handwriting, incomplete pages, and repeat submissions. Each proposed field carries its original page location. Exact reference matching and required-field checks run in code; a model’s guess never becomes a confirmed patient identity or booking. Unclear values stay empty and possible duplicates remain separate until staff review them.
What people approve
The clinic defines required fields, staff access, approved processing locations, and retention rules before connecting documents. Intake staff verify the source, correct extraction errors, and decide whether records belong together. Only authorized staff handle patient documents. The tool does not assess clinical urgency, diagnose a condition, send messages, or book an appointment automatically.
What we’d need to start
- Approved sample referral forms, document layouts, and the intake team’s required-field checklist
- The clinic’s duplicate-review rules, attachment requirements, and appointment request list format
- Permitted document storage and processing arrangements, staff roles, and a review owner
How we’d test it
Use permitted, labeled examples with clear scans, unreadable text, missing pages, different people with similar names, and repeat submissions. Measure field accuracy separately from the number of records processed. Check that possible duplicates never merge automatically, uncertain contact details remain unresolved, and every exported field can be traced to its reviewed source. Compare staff correction effort with the current intake process.
Prepared for Healthcare & wellness. We would agree on the work after reviewing your brief, materials, and requirements.
This is a proposed project, not completed client work. The visuals and numbers show example outputs, not measured results.
