Overview
Our Belief
Burnout isn't about how many hours you work—it's about a lack of believable motivation. We detect early burnout signals and help you accelerate out through pathways that reignite your interests and align with your curiosities.
What we built
- Burnout risk classifier keyed to Maslach Burnout Inventory: emotional exhaustion, depersonalization, personal accomplishment.
- Fine-tuned transformer (Llama 3.1 8B base) trained on our proprietary data set of professional learning and burnout assessments using the most advanced classification techniques.
- Calibrated to Maslach Burnout Inventory dimensions with >90% accuracy vs MBI categories.
- Outputs per-dimension scores plus an overall burnout likelihood with concise rationales.
| Capability | OpenAI Omni Moderation | StageOneGo |
|---|---|---|
| Purpose | Detect harmful content (violence, self-harm, harassment, hate, sexual) | Assess burnout risk and extract actionable insights |
| Burnout Intensity Score | — | Continuous 0-1 score, Pearson r = 0.87 vs MBI |
| Clinical Validation | Policy-based flags for safety violations | Validated on our proprietary data set of professional learning and burnout assessments using the most advanced classification techniques, RMSE = 0.12 |
| Symptom Extraction | — | MBI dimensions: emotional exhaustion, depersonalization, reduced accomplishment |
| Workplace Blocker Detection | Flags sensitive/harmful content | Extracts workload, resource constraints, interpersonal friction, academic strain |
| Trend Analysis | — | Tracks burnout intensity over time: rising, falling, stable |
| Use Case | Content safety and policy enforcement | Employee wellness, early intervention, burnout prevention |
How it works
- Accepts short-form text (journals, notes, chats); light normalization only—no external enrichment.
- Adapter head predicts the three dimensions independently; calibration matches human-rater score distributions.
- Response schema (default):
json
{
"emotionalExhaustion": 0.68,
"depersonalization": 0.31,
"personalAccomplishment": 0.72,
"overallRisk": 0.54,
"rationale": "Primary signals from affective language and depleted tone.",
"modelVersion": "burnout-l3.1-adapter-v2"
}
Boundaries
- Optimized for English, professional-context text; avoid speculative medical conclusions without clinician review.
- Decision support only; not a diagnostic instrument.
- Do not send PII/PHI unless contractually cleared; inputs are not persisted by default.
- Provide ~50 words of de-identified text for reliable outputs.
Validation
- Evaluation results: >90% accuracy vs Maslach Burnout Inventory categories; RMSE 0.12; Pearson r 0.87 for burnout intensity.
- Fast inference: ~200ms p95 latency.
- Outputs include burnout intensity score (0-1) with trend direction, symptoms (emotional exhaustion, depersonalization, reduced accomplishment), and blockers (workload strain, resource gaps, interpersonal friction, academic pressure).