Human–AI Systems · Physiological Computing

Every person has a unique physiological signature. We measure state against their own baseline.

AMI 007 models each person's physiology as a full distribution and measures state against that person's own baseline, capturing the rich patterns that make every individual unique. It is the foundation for AI teammates that know the right moment to speak up, offer support, or let the operator stay focused.

50+Publications, preprints & refereed presentations
LivePhysiological demos at the DAF Modeling & Simulation Summit, 2026
2Best Paper awards
2U.S. provisional patents filed
Published & presented at I/ITSECMODSIM WorldHFESHCI InternationalAHFE / IHSIANNSIMDAFMSAS
What sets us apart

The signal is in the shape, not the level.

Heart-rate variability responds to many things at once: posture, engagement, effort, and fatigue. The pattern of how a person's physiology varies reveals which one is at work.

The traditional approach

One number, one threshold

  • Summarizes physiology as a single value, such as heart rate, HRV, or a "readiness score"
  • Compares it to a population norm or a rolling average
  • Built on group averages and reports levels
  • Well suited to broad, population-level trends
The AMI 007 approach

Your distribution, your baseline

  • Models each person's physiology as a stochastic process, not a single value
  • Measures state as the distance from that person's own reference distributions (rest, productive load, fixation)
  • Captures the full richness of variability, shape, and covariance
  • Gives AI a principled signal for the right moment to engage, with a clear reason why
The mathematics

Built on rigorous mathematics, designed for trust.

Our approach rests on probability theory and optimal transport, the branch of mathematics that measures precisely how one probability distribution differs from another. Rather than asking "Is this number high?", we ask "How far is this moment from this person's own known states, and in which direction?"

How the index works: it measures the least "work" needed to move a person's baseline distribution onto their current one, so it registers changes in shape as well as level. Illustrative animation.
01

Each person is their own reference

A short calibration session captures each person's reference distributions for known states: rest, light load, productive high load, and distraction. Every reading is measured against how this person naturally behaves.

02

The whole distribution, not just the average

We compare full probability distributions using a Wasserstein (optimal-transport) distance. It captures a shift in level and a change in shape, so a change in variability registers even when the average stays the same.

03

A true, interpretable distance

The Wasserstein distance is a genuine mathematical metric that respects the geometry of the signal. Every reading can be explained as how far, and in which way, the present moment sits from a person's known states. That is the foundation for transparent AI.

04

Scaled to the individual

What counts as "close" is set by how much each person naturally varies when nothing has changed. The result is a principled, person-specific scale in place of a one-size-fits-all threshold.

05

Steady in real conditions

Our fusion engine re-weights each sensor second by second according to signal confidence, so the estimate stays steady when a channel becomes noisy or drops out. Task context adds the direction that helps tell productive focus from fixation.

06

Evaluated with discipline

We evaluate models person by person, with cross-validation, confidence intervals, and success criteria set in advance. Partners see rigor, clearly reported.

AccurateSensitive to changes in shape as well as level, where much of the information lives.
ValidTested with real people under controlled, IRB-approved conditions and submitted for peer review.
ExplainableEvery reading traces back to a distance you can inspect and interpret.
UnifiedOne measurement engine and one calibration layer shared across all our systems.
The evidence

The shape of a person's physiology tells the fuller story.

We tested the approach in a large, IRB-approved driving-simulator study, comparing standard heart-rate and HRV measures with our person-referenced distributional index on the same people under the same conditions.

The result points to a simple idea: what separates one mental state from another lies in the shape of a person's physiology, not its level.

Manuscript under review. The method is covered by a filed U.S. provisional patent. Full results will be linked here once they are published.

Finding 1

A richer signal brings personal context to life

A single heart-based number, even when personalized to each driver, captured only part of the picture. The full distribution reveals much more.

Finding 2

The shape of the signal carries the information

On the same drivers, our distributional index consistently detected higher mental demand for most of them, clearly outperforming standard heart-rate and HRV measures.

Finding 3

It tracks what matters over time

The index followed the cycle of support and recovery during the task, and it grew as time on task increased.

Why it matters beyond defense

Personal insight, wherever people and technology work together

When a person's state is expressed in the shape of their physiology, every system that relies on population averages gains a powerful new source of insight. A person-referenced measure can make alerts more meaningful and support more timely across many fields.

Aviation & aircrew

Individual readiness

Workload, fatigue, and disorientation risk in training, read against each aviator's own normal rather than a fleet average.

Space & long-duration missions

Crews of a few

With small, isolated crews, each person becomes their own reference, building a richer picture over months of operation.

Healthcare

Alerts that matter

Person-referenced alerts could help clinicians focus on the moments that truly need attention, easing alarm fatigue.

Transportation

Driver state & handover

Personal baselines could help recognize drowsiness and overload, and support smoother, safer handovers between drivers and semi-autonomous vehicles.

High-risk operations

Control rooms & first responders

For shift workers, operators, and responders, the signal helps distinguish a productive, busy shift from real overload.

Training & education

Adaptive learning

Simulation-based training that adjusts difficulty and feedback to the learner's actual state, so every trainee progresses at their best pace.

Neurodiversity & accessibility

Fair by design

A personal baseline celebrates individual differences, making the technology inclusive from the start.

Mental health & wellbeing

Stress, on your own terms

Tracking stress and recovery relative to a person's own normal, building on our published digital-twin research.

Consumer wearables

Beyond the generic score

Turning data from devices people already own into personal, meaningful insight.

The common thread

Whenever an AI teammate, an alarm, or an interface has to decide whether to interrupt, a signal calibrated to the individual can make that decision more accurate, fairer, and easier to trust.These are exciting directions for the finding. Each application will be developed with its own testing and validation.

Today · Next

What we've built, and where we're headed.

Proven today

  • Real-time fusion engineCardiac and facial fusion that shifts weight per second as signal confidence changes. Evaluated in a large driving-simulator study across hundreds of system-issued interventions.
  • Person-referenced state indexReliably distinguishes high from low mental demand in a controlled laboratory setting.
  • Monitoring and risk-state classificationContinuous individualized state estimation, plus readiness trends across sessions.
  • Live demonstrationsTwo presentations with live BIOPAC physiological demos at the Department of the Air Force Modeling, Simulation & Analytics Summit, 2026.

Where we're headed

  • From the lab to one wearableBringing lab-grade insight to a single wrist signal that people already wear.
  • Fast, confident calibrationDelivering reads with stated confidence from short baseline sessions.
  • Productive focus vs. fixationDistinguishing healthy, productive effort from fixation by the direction of physiological change.

Real-time monitoring you can trust. Our index tracks operator state as it unfolds, providing continuous, individualized monitoring and trend tracking. We report our results with care and precision, so partners know exactly what they are getting.

Technology portfolio

One measurement engine, many missions

Every system builds on the same foundation: an individual physiological baseline.

Fusion

Simulator-tested
Real-time multimodal fusion

A real-time pipeline that fuses ECG with camera-based facial analysis to estimate operator state and issue intervention recommendations, with per-person calibration.

VIAW-HRV

Demonstrated live
Cognitive-state detection

Trimodal ECG/HRV, voice, and facial state detection for human–autonomy teaming. Demonstrated live at DAFMSAS 2026 for pilot–autonomous wingman teaming.

Adaptive Transparency

Human–AI teaming

Interface adaptation driven by cognitive load. It calibrates how much an AI explains to what the operator can absorb, helping sustain appropriate trust.

Physiological Digital Twins

Published
Modeling & simulation

Executable models of individual and crew physiological trajectories for training optimization and pre-deployment evaluation. See the papers →

NA-HAT

Neurodivergent-adaptive teaming

A human–AI teaming framework with individually calibrated support that broadens the pool of qualified operators.

DISORIENT

Concept
Aircrew wearable

A wearable spatial-disorientation recovery aid for aircrew, developed in response to a need aircrew raised.

PRIVATE

Patent pending
Acoustic privacy

Deployable acoustic-privacy capability for sensitive environments. U.S. provisional patent filed.

Peer-reviewed research

Peer-reviewed, published, and open to explore.

50+ peer-reviewed publications, preprints, and refereed presentations. Below is a selection most relevant to our technology; click through to read the papers.

Digital twins

Person-specific models of physiology, stress, and performance.

Engineering mental wellness: A digital twin for chronic stress modeling and real-time intervention
MODSIM World 2025 · Paper No. 77
Best Paper AwardRead paper →
Physiological digital twins for multi-crew training optimization
I/ITSEC 2026 · Orlando, FL
Accepted · Dec 2026
Personalized EEG-driven digital twin models for AI- and VR-based language therapy in children with autism spectrum disorder
MODSIM World 2025 · Paper No. 80
Human digital-twin-style models for early deterioration prediction in austere trauma care: A validation-readiness simulation study
MODSIM World 2026 · Emerging Capabilities Track
Adaptive self-regulation in digital twin therapy: Modeling the development of autonomous emotion regulation skills through human–AI collaborative learning
Lecture Notes in Computer Science (Springer), 2026
A digital twin framework of motor performance variability and fatigue trajectories under cumulative sleep debt
HFES Annual Meeting (ASPIRE 2026)
Accepted

Physiological sensing & human–AI teaming for defense

Operator-state detection, cognitive load, and adaptive training.

Dual-modal stress detection for adaptive autonomous wingman teaming
I/ITSEC 2026 · Presented with live demo at DAFMSAS 2026
Accepted · Dec 2026
Emotion-aware cognitive load management in human–AI teaming for defense operations
MODSIM World 2025 · Paper No. 63
Enhancing decision-making under pressure: Simulating cognitive load in high-stakes training
I/ITSEC 2025
Adaptive simulation-based training for military decision-making: Leveraging IoT-derived cognitive and emotional feedback
MODSIM World 2025 · Paper No. 82
Affective state modeling to predict training dropout in soldier academies
MODSIM World 2025 · Paper No. 64
Emotion-aware autonomous vehicle control: A framework for driver state management through adaptive AI-driven interventions
ANNSIM 2025 · Madrid, Spain
Published

Trust & transparency in AI

How AI can explain itself in ways that strengthen human judgment and trust.

The transparency dilemma: Reconciling trust calibration with perceived autonomy through personality-adaptive AI systems
Lecture Notes in Computer Science (HCII 2026)
Best Paper AwardRead paper →
The transparency paradox in explainable AI: A theory of autonomy depletion through cognitive load
arXiv preprint, 2026
When to trust the machine: A simulation framework for human–AI collaboration
IHSI 2026 · AHFE International
Autonomy in transition: AI, self-identity, and the evolution of human agency
Proceedings of the HFES Annual Meeting, 2025
Ancuta (Anna) Margondai, Founder and CEO of AMI 007
Founder

Ancuta (Anna) Margondai

Founder & CEO · Ph.D. Candidate, Modeling & Simulation, University of Central Florida

Anna combines stochastic mathematics, experimental psychology, and engineering to answer one question: when can an AI system be trusted to act on what it infers about a human?

She designed, built, and validated the real-time fusion system at the core of AMI 007, and brings deep experience running IRB-approved human-subjects research. Before research, she spent roughly two decades in clinical practice.

  • Best Paper Award, HCI International 2026
  • Best Paper Award, MODSIM World 2025
  • Graduate Mentor of the Year, UCF College of Engineering & Computer Science (2026)
  • OpenBCI Galea Discovery Program, 2026 cohort
  • Best Reviewer Award, Simulation (SCS), 2025
Contact

Let's work together

For government program managers, prime contractors looking for teaming partners, and research collaborators.

Point of contact
Ancuta (Anna) Margondai, Founder & CEO
Send us an email