Presage Community

Research & community.

Explore the clinical research behind SmartSpectra and connect with developers building contactless applications.

Connect

Join the community.

Developer updates

SDK releases, API changes, and best practices delivered to your inbox.

No spam. Unsubscribe anytime.

Discord

Ask questions, share projects, and connect with other developers building with SmartSpectra.

Join the Discord

GitHub

Explore open-source SDK samples, integration guides, and developer tools for contactless health monitoring.

Follow on GitHub
Resources

Explore the science.

Algorithm validation detail for each SmartSpectra measurement, including accuracy figures and the conditions each model was tested under.

Relative Arterial Pressure Waveform

Waveform of relative arterial pressure (amplitude in arbitrary units vs. time). Represents waveform shape only (e.g., heartbeat timing, shape and relative changes). Not a blood pressure measurement and cannot be used to estimate systolic or diastolic pressure.

See model card for confidence to error mapping.

Range Valid when pulse rate is 40-110 BPM.

Works when

  • Subject is stationary.
  • Camera is relatively stable; handheld is acceptable.
  • Face and upper chest are visible, unobstructed, and well illuminated.

Does not work when

These conditions may produce inaccurate measurements with high confidence and should be avoided.

  • Large head, body, or camera motion.
  • Chewing gum.
  • Flickering illumination, such as TV screens.
Model card for Relative Arterial Pressure Waveform

Heart Rate Variability

60-second beat-to-beat (NN interval) statistics with confidence. Confidence is 0 until the full 60-second window is reached.

See model card for confidence to error mapping.

Range Valid when pulse rate is 40-110 BPM.

Metrics

  • Mean NN: Average NN interval in milliseconds.
  • RMSSD: Root mean square of successive NN differences in milliseconds.
  • SDNN: Standard deviation of NN intervals in milliseconds.
  • Baevsky Stress Index: A measure of autonomic balance derived from the NN interval distribution, computed as AMo / (2 × Mo × MxDMn). Interpret it relative to a subject's own baseline rather than as an absolute score.

Works when

  • Subject is stationary.
  • Camera is relatively stable; handheld is acceptable.
  • Face and upper chest are visible, unobstructed, and well illuminated.

Does not work when

These conditions may produce inaccurate measurements with high confidence and should be avoided.

  • Large head, body, or camera motion.
  • Chewing gum.
  • Flickering illumination, such as TV screens.
Model card for Heart Rate Variability

Pulse Rate

12-second average pulse rate in beats per minute (BPM) with confidence.

See model card for confidence to error mapping.

Range 40-110 BPM.

Works when

  • Subject is stationary.
  • Camera is relatively stable; handheld is acceptable.
  • Face and upper chest are visible, unobstructed, and well illuminated.

Does not work when

These conditions may produce inaccurate measurements with high confidence and should be avoided.

  • Large head, body, or camera motion.
  • Chewing gum.
  • Flickering illumination, such as TV screens.
Model card for Pulse Rate

Face Analysis

Includes:

  • 478 facial landmark coordinates as x/y pixel locations.
  • Eye blink detection (binary).
  • Talking detection (binary).
  • Facial expression classification with probabilities across anger, contempt, disgust, fear, happiness, sadness, surprise, and neutral.

Works when

  • Face is visible, unobstructed, and well illuminated.

Does not work when

  • Face is significantly rotated and not approximately orthogonal to the camera.
Face landmark numbering reference for Face Analysis

Breathing Rate

30-second average breathing rate in breaths per minute with confidence. Confidence is 0 until the full 30-second window is reached.

See model card for confidence to error mapping.

Range 5-40 breaths per minute.

Works when

  • Subject is stationary.
  • Camera is stable and not handheld.
  • Face and chest are visible and well illuminated.

Does not work when

  • Talking.
  • Large body motion.
  • Unstable camera, including handheld capture.
  • Dark clothing.
  • Tightly striped patterns on clothing causing image aliasing.
Model card for Breathing Rate

Upper and Lower Breathing Waveforms

Upper chest and lower abdominal breathing waveforms (amplitude in arbitrary units vs. time) with confidence. Confidence is derived from breathing rate confidence and remains 0 until 30 seconds is reached.

See model card for confidence to error mapping.

Range Valid for breathing rates of 5-40 breaths per minute.

Works when

  • Subject is stationary.
  • Camera is stable.
  • Face and chest are visible and well illuminated.
  • Lower abdominal waveform requires waistline visibility.

Does not work when

  • Talking.
  • Large body motion.
  • Unstable camera, including handheld capture.
  • Dark clothing.
  • Tightly striped patterns on clothing causing image aliasing.
Model card for Upper and Lower Breathing Waveforms

SDK metrics are offered for general wellness and informational purposes only. SDK metrics have not been cleared by the FDA and may not be used for medical diagnosis or treatment.

Showcase

Built by the community.

Explore developer projects and the latest media coverage.

119 repositories
JavaScript

GeekSafe

Your AI-powered safety mirror: Combining Gemini’s medical intelligence with live biometric sensing to keep you safe from high-risk substance interactions.

sborishchev/geeksafe
TypeScript

MNEMOSYNE

We provide patients with free cognitive assessments and give doctors live access to their patient!

SahandSZH/MNEMOSYNE
Kotlin

Medicus

Streamlining emergency care pathways to prevent unnecessary hospitalizations, reduce system strain, and ensure timely treatment.

Megaman222111/hackcanada-2026
Swift

Behaviourly

Interview Better, Practice Smarter, Land the Job!

allenabraham106/HackCanadaApp
Python

Moha Health - Multi-Agent AI Health Assistant

Moha Health is a multi-agent AI healthcare assistant. Patients describe symptoms via voice or text, and AI agents triage cases, route them to specialists, and generate clinical reports in minutes.

Nafisatibrahim
TypeScript

PosiSense

Real-time AI interview coach that analyzes how you come across on camera (eye contact, expression, and pacing), not just what you say.

TrinhTT8/posi-sense