Technology
An inference engine that shows its work.
This is the page where we use the real words. The HemiSense Array is the sensing hardware and the explainable fuzzy inference engine that reads it, together in one wearable. We call that the brain. HemiAct add-ons are the effectors that act on its output. We call those the body. Products are built by pairing the two.
Layer one
The HemiSense Array
A wearable sensor network at seven sites on the body. Three sit on the midline and four are worn bilaterally, eleven sensors in all. Together they cover the head, the limbs and the trunk, which is what lets one array serve conditions as different from each other as tremor, seizure and gait impairment.
- 1 Headband midline
- 2 Elbow, both arms
- 3 Wrist, both arms
- 4 Chest strap midline
- 5 Waist belt midline
- 6 Knee, both legs
- 7 Ankle, both legs
What the array reads
Neural. Cortical rhythm, event morphology and state, at the head. Motor. Efferent motor signal at the limb sites, and inertial sensing for tremor, bradykinesia, gait and limb use. Autonomic. Cardiac, electrodermal and temperature signals that carry autonomic state and often precede neurological events. No single modality carries every condition, so the array fuses several and is built to keep working when one is unavailable.
Layer two
Why fuzzy logic and not a black box.
Deep networks are excellent at prediction and poor at accountability. In a clinical setting that trade is not neutral, it is disqualifying. Fuzzy inference gives up very little accuracy on graded physiological states and gives back a complete audit trail. The four stages below are the whole mechanism. There is nothing hidden behind them.
Fuzzification
Raw features are mapped onto membership functions that describe degree rather than category. A signal is 0.7 "elevated" and 0.3 "normal" instead of being forced to one side of a cutoff no clinician would defend. The membership functions are the mathematical object in our logo.
Rule evaluation
A rule base built from clinical expertise and refined against data fires in parallel. Each rule is a plain sentence about physiology, which is what makes the whole system legible, and it is what makes adding an indication a matter of writing rules rather than designing hardware.
Aggregation and output
Rule outputs combine into a graded conclusion with an explicit confidence, alongside the specific rules and inputs that drove it.
Explanation
The clinician receives the conclusion and the reasoning in the same view. Disagreement is possible, which is the point. A model that cannot be argued with cannot be trusted.
Architecture
One engine, four functions.
Detect, Monitor, Treat and Predict are four modes of the same inference core, which is what lets a device grow in capability without a new platform underneath it.
Detect
Identify events and states as they occur in daily life. Ambulatory sensing removes the reporting gap between what happens at home and what is described in clinic.
Monitor
Quantify severity over time on a continuous scale, producing a longitudinal record that a twice-yearly assessment cannot approximate.
Treat
Where an intervention is appropriate to the indication, use the same measurements to adapt it to the individual and to the moment. In SureHand the intervention is the out of phase impulse itself.
Predict
Model trajectory and near-term risk, with the confidence and the driving factors stated openly rather than implied.
Configuration
Modular sensors for personalized care.
We do not build a new platform for each condition. We select sites from the array, pair them with a rule base written for that condition, and the rest of the stack stays the same. SureHand uses sites 2 and 3, elbow and wrist, which are the two sites we are building first. A gait product would draw on the waist, knees and ankles. A seizure product would lean on the headband.
Add-ons sit on the other side of the same interface. The array publishes a state estimate with its explanation attached; a HemiAct add-on subscribes to it and acts. That is what makes a counter-tremor sleeve, an adaptive dosing device and a control channel for a third-party prosthetic the same kind of object to the array.
Adherence is a clinical variable. A device that produces perfect data and is left in a drawer produces no data, so the number of sites a product asks the wearer to put on is a design constraint we take seriously.
SureHand in detailWorked example. SureHand: two array sensors, one add-on woven into the sleeve between them, and a rule base written for tremor. Everything else in the stack is shared with whatever we build next.
Research and evidence
What this rests on.
The engine grows out of the founder's doctoral research at the University of Cincinnati in explainable AI, fuzzy logic systems and interpretable fuzzy inference, with peer-reviewed work published through Springer (NAFIPS 2025), the Association for Computational Linguistics (ACL 2026) and the European Conference on Computer Vision (ECCV 2026), and a paper in preparation on explainable fuzzy inference for wearable seizure prediction and neurological monitoring. The company's device intellectual property is separate from university-sponsored work.
We have not yet published on the HemiSense Array or SureHand specifically. When we do, it will be here first, with the data.
The published studies each step of the SureHand loop builds on are laid out, step by step, in how SureHand works.
Validation path
- Now. Engine development and evaluation against public and partner neurological datasets
- Next. Prototype hardware and signal quality testing with research partners under IRB oversight
- Then. Prospective evaluation against accepted clinical reference standards in the lead indication
- Later. Submission pathway determined by indication, claim and risk classification
Commitments
Responsible AI, stated plainly.
- Every clinical output is explainable to the clinician receiving it
- Uncertainty is reported, never hidden behind a single number
- Performance is characterized across populations, not on average alone
- Patient data handling and consent are designed in from the first prototype
- The system supports the clinical decision. It does not make it
Investigational status. HemiLogic devices and software are in development. They are not cleared or approved by the FDA or any other regulatory authority, are not available for sale, and are not intended to diagnose, treat, cure or prevent any disease at this time.
Want the deeper technical overview?
We share architecture and validation detail with investors, clinical collaborators and research partners under NDA.