HealthLens AI (demo) shows how AI can connect a patient's health data into one knowledge bank and serve the patient, the doctor, the hospital and public health, with governance built in.
Demonstration with synthetic data. Not a medical device. Not medical advice.
Loading the demo.
HealthLens AI (demo) is a working example of AI use cases in health, shown together instead of one at a time. It connects one patient's health data into a single knowledge bank. The same records then serve four audiences: the patient, the doctor, the hospital and the public health authority. A governance view shows the controls that keep each AI output accountable.
Everything runs in your browser. The data is synthetic and generated in code. The demo calls no language model and sends no data anywhere. It is a demonstration, not a medical device, and it gives no medical advice.
The data model follows the shape of HL7 FHIR resources and uses LOINC and ICD-10 codes. A fixed random seed generates four fictional patients, one hospital and 52 weeks of surveillance data, so every visitor sees the same records. The assistant and the brief use keyword retrieval and templates, not a language model. The interface works in English and Arabic, with right-to-left layout and light and dark themes.
The governance view follows the six principles of the World Health Organization guidance Ethics and governance of artificial intelligence for health (2021).
The demo states its limits. The safety rules of the assistant are phrase patterns, and they miss some phrasings. The governance view reports the measured results. No clinician has validated the medical content.
HealthLens AI is a demonstration. It shows one patient knowledge bank serving four audiences: the patient, the doctor, the hospital and the public health authority. A governance view shows the controls behind each AI output. It is not a medical device and it does not give medical advice.
No. All data is synthetic and generated in code from a fixed random seed, so every visitor sees the same fictional patients, hospital and countries. The demo runs in your browser and sends no data anywhere.
No. The assistant and the doctor brief use keyword retrieval over approved content and the synthetic record, then fill fixed templates. This keeps the demo free of external AI calls and makes every statement traceable to a source. A production system could place a language model behind the same safety and citation layer.
No. The medical content is general, conservative and labelled as illustrative. No clinician has validated it. For any health concern, speak to a qualified health professional. For warning signs such as chest pain or trouble breathing, seek urgent care.
In this demo, a patient knowledge bank is one structured record of a patient’s health data: conditions, medicines, laboratory results, home readings and visits. Each user sees the part that fits their role, and the patient controls what is shared. A production system would typically exchange such data using a standard such as HL7 FHIR.
A model card is a short document that states what a model is for, what it must not be used for, what data it relies on, how it performs and where it fails. It helps clinicians, managers and reviewers judge whether the model fits their setting. The governance view of the demo contains a worked example.
The World Health Organization set out six principles in 2021: protect human autonomy; promote human well-being, safety and the public interest; ensure transparency, explainability and intelligibility; foster responsibility and accountability; ensure inclusiveness and equity; and promote AI that is responsive and sustainable. The governance view maps each principle to a control in the demo.
Each district is compared with its own recent history. A week raises a signal when the case count is above the mean of the previous 12 weeks plus two standard deviations, and is at least 10 cases. A signal is a prompt for an epidemiologist to investigate. It is not a confirmed outbreak, and a duty officer must verify it before any advisory reaches patients.
Every AI statement carries a numbered citation that opens the source record. The doctor accepts, edits or rejects each section of the brief. An audit log records each action, a lineage view traces each figure to its source, and an equity monitor compares results across groups.
Built by Shahzad Asghar, who designs and governs data and AI systems for public-sector and humanitarian settings. More work: applied AI projects.