AI interpretation
Interpretation is an optional layer after metrics exist. Every output cites computed values from completed runs — the LLM does not invent numbers.
Prerequisites
- Completed HRV, cohort summary, baroreflex, or related pipeline run
- Ollama credentials configured in
backend-analysis(see repo rootai.md)
Descriptive mode
Plain-language summary of what the metrics suggest physiologically.
Available for:
- Single HRV run (includes SQI and arrhythmia burden when present)
- Cohort comparison (group differences with p-values and group means)
- Subject hub (summary across available modalities)
Where to launch:
- Run detail page → Interpret
- Cohort workspace → Interpret
- Subject hub → Interpret
Mechanistic mode
Cross-domain hypotheses when multiple metric domains are present on a subject:
- Autonomic (HRV)
- Hemodynamic (BP variability, dipping)
- Baroreflex sensitivity
- Respiration / RSA
- Rhythm burden
Outputs include pathway tags (for example autonomic, hemodynamic), confidence levels, and linked metric badges. Falls back to descriptive mode if only one domain is available.
Where to launch:
- Subject hub → Interpret (mechanistic)
- Run detail (when multimodal context exists)
- Cohort outlier panel
Guardrails
Every interpretation artifact includes:
- Research / non-clinical disclaimer
- Metric citation validation (numbers must match run artifacts)
- Automatic caveats for low SQI, high artifact rate, or missing modalities
Copy text or export JSON from the interpret panel.
What interpretation does not do
- Diagnose conditions or recommend treatment
- Replace expert review of signal quality
- Compute new metrics — it explains metrics already in artifacts