South Africa’s Open Source AI Scribing Project: Early Lessons from the Frontline
- May 7
- 3 min read
We’re learning fast by testing early prototypes with clinicians across the Western Cape to understand what works, what doesn’t, and what “good” looks like in real facilities.
The Primary Health Care (PHC) Optimisation project is developing and testing an open-source ambient AI scribing tool in Western Cape public health facilities—designed to reduce documentation workload, support better-quality clinical notes, and help clinicians spend more time with patients. The initiative is a collaboration between Jembi (technical lead) and the Western Cape Department of Health and Wellness (WCDHW).
Site visits across four facilities in both the Cape Town metro and rural areas revealed strong clinician interest and surfaced critical insights that have shaped successive prototype iterations.
This Project at a Glance
Where: Western Cape public health facilities
Who: Frontline primary care clinicians, facility managers, provincial and city health departments
What: An open-source ambient AI scribing tool that integrates into existing clinical systems, including SPV (Single Patient Viewer), CIS (Clinical Information System), and HECTIS—fitting into established workflows rather than disrupting them.
When: We’re in a rapid prototyping phase now—designing, building, and testing in short learning loops. Next is the pilot phase, where the tool will be trialled in real-world settings, with clinicians and patients, following confirmed protocols and approvals.
Why: Reduce the documentation burden so clinicians can spend more time on care, shorten the overall consultation duration, and strengthen the quality of clinical documentation.
What we're learning
Clinicians spend up to two hours a day on paperwork. Time that can come at the cost of patient care and contribute to burnout as clinicians try to juggle everything.
Through hands-on demonstrations and mock consultations at facilities including Mitchells Plain CHC, St Vincent's CDC, District Six CDC, and Worcester CDC, the team is learning exactly what frontline care requires.
Clinicians are leaning in. One experienced AI scribe user called the tool "essential, not a nice-to-have" and noted that local language support would be "a game-changer". Facility management at one site, initially wary of yet another pilot, said after the demo: "This will be the best pilot yet!"
Real-world workflows are complex. Doctor-patient consults often include multiple speakers, frequent interruptions, and off-record moments. Our tool must handle Afrikaans, isiXhosa, English, and code-switching, the natural mixing of languages mid-conversation. This approach also allows the clinician to maintain complete control over reviewing, editing, and approving every note before saving it.
Integration matters more than features. The partners made it clear that they didn't want yet another system or app for clinicians to use. So the team is designing a lightweight plugin that works inside existing host applications like SPV, CIS, and HECTIS. Fitting into their established workflows rather than disrupting them.
SPV’s project lead reported in a March workshop that SPV’s e-scripting module already serves over 1,000 daily prescribers and has processed more than 1.2 million scripts and five million line items—an indicator of how routinely SPV is used in frontline services. This prescription workflow is where the AI scribe will be introduced first, making tight, low-friction integration essential to accelerate an already working process.
The AI scribe is being built as open-source software under a permissive licence (MIT or MPL 2.0), so health departments can download, use, and adapt it without vendor lock-in. There are no ongoing licensing fees, but implementation and support still need resourcing.
Designing with privacy and responsibility at the center
From the start, the project has prioritised informed consent pathways, minimal data retention, and human-in-the-loop validation. Clinicians must approve AI-generated notes before they become part of the medical record. The scribe-generated SOAP note is editable within the plugin before it passes to the host system, keeping the clinician in full control of what is saved. We are having discussions on consent pathways, audit trails, and data sovereignty alongside technical design to ensure both legal clarity and clinician confidence.
The team is also tracking key risks, including data residency, model performance in local languages, and the danger of “rubber-stamping” AI notes without adequate review.
Collaborating to get it right
This work is only possible because of close collaboration with the Western Cape Department of Health, the City of Cape Town, and facility teams who are generously sharing their time and expertise. Aurum Institute is leading evaluation studies while Reach Digital Health is developing patient-facing functionality.
The ambient scribe sits within a broader solution (including AI-generated artefacts and downstream clinical documentation) that is being co-developed with the WCDHW's Provincial Health Data Centre (PHDC), and we are still confirming the full scope.
With the prototype under review and two key workshops completed in March 2026, the team is preparing for the pilot phase, pending confirmation of scope, protocol, and approvals.
