Early triage decisions shouldn’t guess.
Safety-first clinical decision support for primary care and telemedicine — built to know its limits, not overrule clinicians.
What the system actually does — in 5 seconds
Most AI answers questions.
Aetherion answers when not to trust itself.
Standard models force a prediction even on data they've never seen, leading to confident hallucinations.
Aetherion measures distance from its training distribution. If it doesn't know, it says so.
The front door of care is high-stakes—and often ambiguous.
Symptom Overlap
Early symptoms overlap across many conditions. Distinguishing a benign viral illness from a serious infection is statistically difficult without labs.
Time Pressure
Primary care and telemedicine visits are short. Limited context and high volume drive variability in decision-making and documentation.
Triage Risks
Under-triage risks delayed care for critical patients; over-triage drives unnecessary ER visits, costs, and system overload.
For Clinicians
- Faster risk prioritization
- Clear uncertainty boundaries
For Health Systems
- More consistent front-door decisions
- Audit-ready triage rationale
For Pilots & Mentors
- Evaluation-ready prototype
- Scope & limits stated upfront
Decision support that prioritizes safety and clarity.
What Aetherion Is
- Workflow-integrated clinical decision support.
- A "second opinion" for triage ranking.
- A safety layer for high-risk/uncertain cases.
What Aetherion Is Not
- Autonomous diagnostic AI.
- A finalized medical device.
- A replacement for clinical protocols or physician judgment.
Provides a ranked list of potential conditions with calibrated likelihoods to support clinical reasoning.
Categorizes cases from Low to Critical urgency, aligning with standard triage protocols.
Deterministic rule layer catches "cannot miss" signs regardless of AI model output.
explicitly warns when a case is uncertain or Out-Of-Distribution (OOD), preventing silent failures.
Where we are today
Evidence focuses on reliability signals (calibration + OOD behavior), not clinical outcome claims.
Conditions Covered
High-volume primary care focus
Top-1 Accuracy
On validation set
Macro-F1 Score
Balanced performance
Calibration (ECE)
Reliable probability estimates
Working API + UI
Interactive prototype
Validation Roadmap
Phase 1: Architecture & Prototyping (Current)
Developed core pipeline, trained on synthetic datasets (derived from verified medical ontologies), built interpretable UI, and implemented OOD safety gating.
Phase 2: Clinician-in-the-Loop Evaluation (Next)
Qualitative study with primary care physicians to assess workflow fit, explanation utility, and trust calibration.
Target: 10–15 clinicians reviewing simulated triage cases.Starting Q2 2026
Phase 3: Retrospective Validation
Rigorous testing on de-identified real-world datasets from partner clinics to benchmark against standard care.
How it works
Minimal Guided Intake
Clinician or intake form collects key symptoms, duration, and vitals. Designed for speed.
Red-Flag Safety Layer
Deterministic rules scan for critical signs (e.g., crushing chest pain) to trigger immediate escalation.
Interpretable Model
Interpretable tabular model (LightGBM) estimates probabilities. No black-box complexity.
Reliability Check
System runs calibration verification and Out-Of-Distribution (OOD) check.
Clinician Output
Results presented with confidence intervals and 'Why' explanations. Full audit logging.
Built to fail safely.
Decision Support
Designed as decision support, not autonomous diagnosis.
Conservative Escalation
Prioritizes conservative escalation when uncertainty is high.
OOD Warnings
Explicit warnings for out-of-scope or out-of-distribution cases.
Clinician-Pulled
Clinician-pulled workflow reduces alert fatigue.
Why this is different
Static, rule-heavy, slow to adapt.
Often rigid for complex presentations.
Noisy, interruptive, high fatigue.
Integrated but often ignored.
Strong accuracy, weak safety signaling.
Poor at handling uncertainty.
Aetherion exists between protocols and black-box AI — prioritizing safe next steps over confident guesses.
Innovation Video
Transcript Highlight
"...We built Aetherion to solve the 'front door' problem in healthcare. It knows when to speak up—and more importantly, when to stay silent if the data isn't clear."
The Team
End-to-end system design, ML pipeline, API/UI, and safety orchestration.
Medical guideline mapping, benchmarking, and clinician-facing UX.
Architecture review, scalability guidance, and engineering mentorship.