Five models read every ECG. Only an agreed reading ships.
AI ECG interpretation for insurers, hospitals, clinics and labs. A consensus engine, a confidence score printed on the report, and a human when the models disagree.
Not one model with an accuracy number. Five, made to agree.
A single model produces a confident answer whether or not it should. The architecture below is designed so that uncertainty surfaces instead of being hidden inside an average.
Five models, read independently
The same twelve lead trace goes to five models that do not see each other's output. Independence is the point: agreement only means something when it was not coordinated.
Adaptive multi model voting
A consensus engine weighs the readings against each other rather than taking a majority. Only an agreed reading becomes a report.
Anomaly cross checking
Findings are cross checked against the rest of the trace and against the patient's other results, so a single odd lead does not carry a diagnosis on its own.
Disagreement escalates to a person
Where the models do not agree, the case goes to a doctor. It is not averaged, not rounded, and not shipped with a hedge in the wording.
The confidence score is printed on the report
Every reading carries the score it was made with, so the person acting on it knows how much weight it holds. A reading the system was unsure about does not arrive looking identical to one it was certain about.
Screening counts come from the Cardiotrack screening database. Physician sign off is included on the insurance and device plans.
Numbers your actuary can argue with. Which is the point.
Measured on production cases rather than a demo set, and stated with the benchmark beside them.
93% sensitivity
AI ECG interpretation, against a published benchmark range of 80 to 90 percent.
95% specificity
Against a published benchmark range of 85 to 95 percent.
Under 60 seconds
From the last beat to the interpretation, so a poor trace is recaptured before the technician leaves the house.
500,000 plus ECGs
The size of the South Asian training set behind the model.
Cardiotrack figures are internal benchmarks measured on production cases. An independent validation study is due in the third quarter of 2026. The device technology has been reviewed by Swiss Re, Munich Re and Reinsurance Group of America.
The same engine. Bought differently by each of you.
Insurers integrate it. Hospitals and clinics use it per scan. Labs get it bundled with the device. Pick the door that matches how you work.
Insurers, over the API
Send a trace, receive a structured, physician signed reading with its confidence score, inside your own underwriting workflow. Built for volume and for the audit trail your regulator will ask about.
Hospitals and clinics, per scan
Cardiologist grade interpretation on demand, without a cardiologist on the floor. Use it on the ECG machines you already own, priced per scan rather than per seat.
Labs, bundled with the device
The interpretation comes with the Cardiotrack device and the platform plan, so a screening you run at a camp or in a home comes back read and signed without a second vendor.
Patent pending, and looked at by people who pay for being wrong.
Validated AI ECG interpretation
Five models read the same ECG independently, a consensus engine weighs them against each other, and only an agreed reading becomes a report. Where the models disagree, the case is escalated to a person instead of averaged into a number.
Reviewed by
Three of the world's largest reinsurers have reviewed the device technology.
Recognised by
No ECG machine? Ours is the complementary half
The interpretation runs on traces from the machines you already have. If you do not have one, or you need to screen away from the clinic, the Cardiotrack 12 lead device records, uploads over Bluetooth and returns the same reading from a home, a camp or a village. The AI is the product. The device is how you reach the patient when the patient cannot reach you.
The AI drafts. A doctor decides.
On the insurance and device plans, a physician signs the reading before it goes anywhere. The AI never signs its own work.
Ten doctors on call
Pickup in under 15 seconds during working hours, so the sign off is not the thing that slows the report down.
The draft is not the verdict
The AI proposes a reading and its confidence. The doctor accepts, edits or overrules it, and the version that ships is the doctor's.
Both are on the record
Every read is logged with its confidence score and the physician's final call, so an audit can see where the two agreed and where they did not.
One compliance boundary, end to end
The trace is captured, read, signed and delivered inside the Cardiotrack ecosystem. Personal and health data is resident in India on secure cloud infrastructure, access is set by role, every case carries an audit trail, and access and deletion requests are handled in line with the DPDP Act 2023.
The questions we actually get.
Can it read ECGs from our existing machines, or only the Cardiotrack device?
From your existing machines. The interpretation works on a standard twelve lead trace, whichever machine recorded it. The Cardiotrack device exists for screening away from the clinic, not as a condition of using the AI.
How is it priced: per scan, subscription or API?
All three, depending on how you buy. Hospitals and clinics usually take it per scan, insurers take it as an API against volume, and labs get it inside the device and platform plan. We have not published a rate card because the per scan price moves with volume. Tell us your monthly ECG count and we will give you a number rather than a range.
What happens when the models disagree?
The case escalates to a doctor. It is not averaged, not resolved by majority vote, and not shipped with hedged wording. Disagreement is treated as information, which is the whole reason five models read the trace instead of one.
Is this regulated as medical device software under CDSCO rules?
The Cardiotrack 12 lead device is CDSCO licensed. The interpretation software sits alongside it, and its regulatory position depends on how it is deployed and by whom. If you need the specifics for your own compliance file, ask and we will put you in front of the person who can answer it properly rather than summarise it here.
Can we run a validation pilot on our own historical ECGs?
Yes, and it is the way we would rather be judged. Send a set of traces you already have signed answers for, and we will return our readings with their confidence scores so you can score us against your own ground truth instead of ours.
Send us an ECG. See what comes back.
The fastest way to judge an interpretation engine is to give it a trace you already know the answer to. Tell us whether you are an insurer, a hospital, a clinic or a lab, and we will set that up.