HMIE · H-Link Medical Intelligent Engine
HMIE, the H-Link Medical Intelligent Engine, runs open medical models on servers we control. Each feature does one task, shows its sources where it has them, and leaves the decision to the clinician. Where a number can be calculated, it is calculated, not guessed by a language model.
Two things that can carry clinical text do leave our servers: questions typed into the knowledge assistant go to Google to find matching sources, and the live dictation preview uses your browser’s own speech service (Google in Chrome, Microsoft in Edge). Stock and catalogue tools, which carry no patient data, can also use cloud AI.
The problem
A medical officer in a district hospital is often the most senior clinician for miles, with no one to ask.
Time spent writing notes and coding diagnoses is time not spent with patients.
A confident answer with no source is dangerous in medicine. Clinicians need to see why.
Capabilities
Suggestions a clinician can review, accept or discard — never an automatic decision.
Answers drawn from medical literature, with each statement linked to the source it came from.
Less typing, more consulting.
The numbers are calculated with statistics, not guessed by a language model.
A second read of an NHIS claim before it goes out, claim by claim.
Workflow
The clinician dictates, asks a question or opens a chart.
HMIE drafts a note, a differential or a code.
Knowledge answers carry numbered citations to the published source.
The clinician accepts, edits or discards it.
Most decision-support calls are logged with the model version that answered.
The H-Link Medical Intelligent Engine: the layer that runs H-Link’s clinical AI. It uses open medical models, in a small and a large size, on servers we control, and each feature calls it for one task.
No. Every feature presents a suggestion for a clinician to review. The clinician decides, and remains responsible for the decision. None of these features has regulatory approval as a medical device.
Patient-data AI features run on HMIE, on GPU servers we control at a hosting provider named in our sub-processor list. A guard in the software refuses to send those requests to cloud AI companies, and an automated check blocks new code that connects to them directly. Two things that can carry clinical text do leave our servers: questions typed into the knowledge assistant go to Google to find matching sources, and the live dictation preview uses your browser’s own speech service (Google in Chrome, Microsoft in Edge). Stock and catalogue tools, which carry no patient data, can also use cloud AI.
In some features, yes, and only on the terms we agree with your facility. Screening images are kept for model improvement only when the patient’s consent is recorded, and clinicians’ corrections to AI suggestions are used to evaluate and improve those features. We agree how your facility’s data may be used before go-live.
Tell us how your facility runs today. We will show you the platform on the departments you care about, and scope a pilot with the numbers agreed up front.
