evidenceos-aya

Aya — the AI-native clinical reasoning assistant at the point of care. This repo contains the public specification, open methodology references, and integration documentation for Aya v1.

Status

Active development. Aya v1 targets facility-tier T0–T2 deployment (district hospital through tertiary regional).

What's here

• spec/ — Public product specification (what Aya does, what it does not)
• methodology/ — Reasoning framework references (CBI-M alignment, ALERT-TBI, GRADE integration)
• integration/ — FHIR R4 integration guide + DHIS2 connector spec
• examples/ — Sample clinical encounter flows with decision trace

What Aya does

Aya is a bedside clinical decision support assistant for traumatic brain injury and acute neurological care. It:

• Accepts structured clinical input (GCS, vital signs, mechanism of injury, available biomarkers)
• Returns evidence-graded recommendations with explicit uncertainty bounds (conformal prediction sets)
• Classifies patients using the NINDS CBI-M framework (CBIM-01 to CBIM-16)
• Adapts recommendations to facility capability tier (T0–T3) and physiological context (T×M×P)
• Logs every recommendation as an immutable Evidence Capsule with full provenance

What Aya does NOT do

• Aya does not make autonomous treatment decisions — it surfaces evidence for clinician review
• Aya does not access patient data without explicit consent and FHIR/DHIS2 integration configured by the institution
• Aya does not provide recommendations outside its validated scope (TBI + acute neurological); it returns ABSTAIN for out-of-scope queries

How this fits the broader EvidenceOS Group

Aya is the bedside interface of the EvidenceOS evidence-to-bedside stack. It sits at Layer 8 of the 10-layer architecture — consuming Evidence Capsules generated by the upstream synthesis and multiverse analysis layers. Aya's recommendations are traceable, via Evidence Capsule provenance chains, to their source systematic review records and individual study citations.

See: evidence-commons for the canonical evidence layer | evidence-capsules for the capsule schema.

Contributing

Clinical decision support AI requires rigorous validation before deployment. We welcome:

• Bug reports and edge-case documentation from clinical users
• Translations of the clinical reasoning documentation
• Facility-tier implementation reports

We do NOT accept contributions that modify the recommendation logic without passing the full validation suite.

See CONTRIBUTING.md for the contribution and validation pathway.

License

• Specification and documentation: CC-BY-4.0
• Integration code: MIT
• Clinical validation datasets: not open (IRB-governed)

Maintainer

EvidenceOS Research Team — research@evidenceos.com
Clinical Safety queries: safety@evidenceos.com

Related repos

• evidenceos-bridge-tbi — BRIDGE-TBI deployment framework
• evidence-capsules — Evidence Capsule schema
• cbim-framework — CBI-M patient classification ontology
• evidenceos-lab-in-a-box — Institutional deployment kit