Research-Validated
Core Engine
We discovered that in real-world conditions, the AI model doesn't matter — labels do. A.R.I.A. builds on that finding: a calibration methodology that adds +10 percentage points over generic wearable models in controlled conditions. The missing layer between raw biosignals and actionable insight.
Sensor Layer
Samsung Galaxy Watch · Empatica · Oura Ring · Any Wearable
A.R.I.A. Standardization Layer
Input Adapter
(Any Sensor)
Edge Processing
(On-Device AI)
Standardized Format
(Unified Schema)
Application Layer
Therapy · XR · Productivity · Automotive · Smart Environments
Explainable by Design
A.R.I.A. uses explainable ML with feature attribution: every classification traces back to specific physiological features — which HRV component shifted, which EDA pattern triggered, what the accelerometer ruled out.
The system doesn't just tell you thatyou're stressed — it surfaces the physiological evidence behind the classification. Designed to connect these signals to real-world context as the platform matures.
No black box. No opaque scores. Every output is grounded in transparent feature attribution that enterprises can audit and trust.
Privacy as a Prerequisite,
Not a Feature
A.R.I.A.'s production architecture is designed with privacy as a structural guarantee — not a policy add-on. Current research operates under IRB-approved, GDPR-compliant protocols.
On-Device Processing
Designed for all processing to run locally on your phone or watch. Raw biosignals never leave the device.
AES-256 Encryption
Data encrypted at rest. Designed so that zero raw biometric data exits the user's device.
Compliant by Architecture
Architected for GDPR and HIPAA compliance by design — not by policy. Edge-native means no cloud data exposure.
Zero-Trust Model
Edge-native architecture designed so that compliance is built into the system, not bolted on.
Research-Validated Core Engine
A lightweight, cross-platform runtime performing real-time, edge-only conversion of raw biometrics into standardized A.R.I.A. data packets.
Target Platforms (Phase 2+)
Android / Wear OS
iOS
Web
REST API
Hardware Integrations
Phase 1 Primary
Samsung Galaxy Watch 8
Validated — 3-channel NormWear fine-tuning shows <1pp accuracy loss versus 4-channel research devices. Wear OS + phone companion app complete. POST /v1/stream and WebSocket /v1/ws for live sensor data → prediction pipeline.
Previous Work
Muse S EEG Headband
Full-protocol pipeline for real-time EEG signal processing. Used in live installations at Sónar+D and Science Week.
Roadmap
Garmin / Oura / WHOOP
Consumer wearable integration via HRV-spectral analysis. Apple Watch blocked by raw sensor restrictions.
Adaptive Personalization Tiers
Tier 1 — Zero-shot
Works immediately using population-level models. Comparable to existing wearable baselines.
Tier 2 — Guided Calibration
A 5–15 minute onboarding session personalizes the model. Lab evidence: +10pp accuracy gain over zero-shot.
Tier 3 — Passive Personalization
Continuous improvement over weeks via self-report integration. No friction after onboarding.
State-Agnostic by Design
The calibration pipeline doesn't know what state it's calibrating — it personalizes whatever the labels describe. Today that's arousal (stressed/neutral/calm). Tomorrow it's focus, fatigue, or any cognitive state measurable from wearable sensors. This architectural decision means every new sensor integration and every new label type expands the platform without rebuilding the core.
| State | Signal | Status |
|---|---|---|
| Arousal (3-class) | Wrist PPG + EDA + ACC | Validated (89.4% lab) |
| Emotion (5-class) | Wrist + voice | Layer 2 roadmap |
| Focus / attention | EEG (Muse S) | Integration proven (Sónar+D, Science Week) |
| Fatigue / energy | Longitudinal wrist HRV | Phase 1 field data |
| Custom states | Any labeled moments | Layer 3 embedding space |
Works on Your Existing Watch
A.R.I.A.'s Adaptive Tiered Model is designed to adapt across different sensor capabilities — from research-grade wristbands to consumer smartwatches. No new hardware purchase needed.
Full Sensor Suite
89.4% Accuracy
EDA + HRV + accelerometry with guided calibration. Lab-validated balanced accuracy.
Consumer Wearables
Samsung Validated
Samsung Galaxy Watch: 71.8% balanced accuracy, <1pp loss versus research-grade devices. Garmin, Oura, WHOOP on roadmap.
Compatible with 560M+ devices already in circulation worldwide.
Integration
A.R.I.A. is built as an API-first platform — designed to sit between any wearable sensor and any application that needs to understand human state.
Today
FastAPI with 6 endpoints including real-time streaming via WebSocket. 416 automated tests. Samsung Galaxy Watch validated. Research partners can run our calibration analysis on their own data.
Roadmap
B2B SDK for wearable OEMs and wellness platforms. Unified sensor abstraction layer supporting Samsung, Garmin, Oura, WHOOP.
Paper submitting to ACM IMWUT, May 2026. Field validation experiment starting June 2026.
For Researchers
Run our calibration analysis on your wearable dataset — at no cost. Get per-subject accuracy with and without personalization, calibration gain analysis, and architecture comparison.
Learn MoreFor Device Makers
We have the only published calibration benchmark on Samsung Galaxy Watch data. Our paper measures exactly what it costs to personalize — and how to do it.
Get in TouchIntellectual Property
The moat is the calibration methodology. No prior work measures calibration cost across architectures — we're the first, and the paper (ACM IMWUT, submission May 2026) establishes the standard. Every researcher who builds on this work uses our methodology as the benchmark.
Calibration Methodology
First Systematic Calibration Cost Measurement
Peer-reviewed paper targeting ACM IMWUT. No prior work compares calibration speed across model families under identical protocols.
Patent Filed
Quantum-Inspired Emotional State Mapping
Provisional patent application for core methodology.
Trademarks
“Human-State Intelligence API” (US pending)
“KUMA” (therapeutic application)
“Information Abstraction Layer” (core concept)