Shifa's intake team was drowning in unstructured patient referrals across 12 clinics in Jeddah. Their internal data team was 6 months booked.
The squad
A staff ML engineer and a data engineer, both with healthcare backgrounds and Arabic NLP experience. Live in 9 days from kickoff.
How we worked
- Weeks 1-2: Audit existing data, define the labels, write a privacy-safe sample contract aligned with NPHIES + MoH guidelines.
- Weeks 3-5: Train the first triage model on de-identified Arabic intake notes; ship the eval harness.
- Weeks 6-7: Wire it into the Shifa intake app behind a clinician override.
- Week 8: Production rollout with a 20% safety margin and a rollback plan.
Outcome
- 70% reduction in unhandled intake at end-of-week.
- Clinician override rate stabilized at 8% — well below the agreed threshold.
- Full audit trail meeting Saudi Ministry of Health requirements from day one.