Assessment Methodology
Overview
Mind Measure employs a multimodal assessment approach combining audio, visual, and text analysis to generate wellness scores. The platform supports two assessment types:
- Baseline Assessments: Establish personal baseline using validated clinical questions (PHQ-2, GAD-2, Mood Scale)
- Daily Check-ins: Brief conversational wellness monitoring
Current Implementation
For detailed technical documentation, see the Assessment Engine section, which covers:
| Component | Documentation |
|---|---|
| System Architecture | Assessment Engine Overview |
| Baseline Pipeline | Baseline Assessment |
| Check-in Pipeline | Daily Check-ins |
| Audio Analysis | Audio Features |
| Visual Analysis | Visual Features |
| Text Analysis | Text Analysis (Bedrock) |
| Score Fusion | Scoring Algorithm |
Key Technologies
| Component | Technology |
|---|---|
| Conversational AI | ElevenLabs React SDK |
| Text Analysis | AWS Bedrock (Claude 3 Haiku) |
| Visual Analysis | AWS Rekognition |
| Audio Analysis | Client-side Web Audio API |
| Database | Aurora PostgreSQL |
Scoring Summary
V2 Scoring (December 2025)
Daily Check-ins:
- 70% Text (Bedrock analysis)
- 15% Audio (voice features)
- 15% Visual (facial features)
- Sanity floor: 60 minimum for positive check-ins
Baseline Assessments:
- 70% Clinical (PHQ-2 + GAD-2 + Mood)
- 15% Audio
- 15% Visual
Validated Scales
PHQ-2 (Depression Screening)
- 2 questions, scored 0-6
- Score ≥3 indicates possible depression (positive screen)
GAD-2 (Anxiety Screening)
- 2 questions, scored 0-6
- Score ≥3 indicates possible anxiety (positive screen)
Mood Scale
- Single question: "Where would you put your mood on a scale of 1 to 10?"
- Direct self-report of current emotional state
Research Foundation
Evidence Base
The multimodal approach is grounded in research on:
- Vocal biomarkers: Pitch, speaking rate, and voice quality correlate with depression/anxiety
- Facial expression: Smile frequency, eye contact associated with positive affect
- Linguistic markers: Word choice and sentiment indicate emotional state
Validation Status
| Aspect | Status |
|---|---|
| PHQ-2/GAD-2 | Validated clinical instruments |
| Audio features | Research-based, not clinically validated |
| Visual features | Research-based, not clinically validated |
| Fusion weights | Initial calibration, pending validation |
Research Priorities
- Longitudinal validation: Track scores against outcomes
- Clinical correlation: Compare with PHQ-9, GAD-7
- Population norms: Establish baseline distributions
- Weight optimisation: Data-driven calibration
Limitations
Technical
- Browser/device variability affects audio/video quality
- Requires stable internet for AI processing
- No control over recording environment
Clinical
- Not diagnostic: Monitoring tool, not clinical diagnosis
- Professional oversight: Requires clinical interpretation
- Crisis detection: May miss acute mental health crises
Ethical Considerations
Data Privacy
- Minimal data retention
- Encrypted in transit and at rest
- User consent required
- No data shared with third parties
Algorithmic Fairness
- Diverse population validation needed
- Regular algorithm auditing
- Transparent scoring methods
For full technical details, see the Assessment Engine documentation.
Last Updated: December 2025