Building the operating system for sovereign health intelligence.
We automate the production of trustworthy clinical evidence for health AI systems — evaluation, safety, and governance infrastructure with mathematical assurances. Starting with neurodiagnostic AI (TBI first vertical), expanding to every domain where clinical AI needs to prove it works before it deploys.
"82% of clinicians know about TBI blood biomarkers. 6% use them. The gap isn't science — it's evidence infrastructure."
Open-source research packages powering clinical validation intelligence. Built on evidence from $165M+ of publicly funded research (TRACK-TBI, CENTER-TBI, InTBIR).
| Package | What It Does | Language |
|---|---|---|
| evidenceos-multiverse | 12-axis specification curve analysis — systematically tests model robustness across thousands of defensible analytical choices | Python |
| evidenceos-conformal | Distribution-free prediction intervals (MondrianCQR) — models know when they don't know | Python |
| evidenceos-knowledge | Knowledge graph reasoning (GraphRAG + DGCI) — explainable clinical evidence synthesis | Python |
| evidenceos-tabular | Clinical prediction model pipeline — loading, harmonization, and feature engineering for TBI datasets | Python |
| evidenceos-ontology | NINDS TBI Common Data Elements — 13 machine-readable YAML schemas (7,780 lines) | YAML |
| evidenceos-acquisition | Evidence acquisition pipeline — automated search, screening, and knowledge graph construction | Python |
| evidenceos-trajectories | Neural ODE biomarker trajectory modeling — temporal disease progression | Python |
| evidenceos-living-review | Living systematic review automation — continuous evidence synthesis | Python |
| evidenceos-bench | Clinical AI benchmark suite — TRIPOD+AI compliance and model comparison | Python |
All 10 packages: evidenceos-research (Apache-2.0)
Methodological foundations: Specification curve analysis (Simonsohn et al. 2020), conformal prediction (Vovk et al. 2005), TRIPOD+AI (Collins et al. 2024), Riley sample size criteria (Riley et al. 2020).
Four integrated product families, one evidence foundation. Each product draws from the same Canonical Evidence Library (CEL) — versioned, DOI-addressable analytical artifacts that serve as the evidentiary ground truth.
┌─────────────────────────┐
│ Evidence Foundation │
│ (CEL + Knowledge Graph │
│ + Ontology + LSRs) │
└────────┬────────────────┘
│
┌─────────────────────┼─────────────────────┐
│ │ │
┌────────▼────────┐ ┌────────▼────────┐ ┌────────▼────────┐
│ Clinic-in-a-Box │ │ Lab-in-a-Box │ │University-in-Box│
│ │ │ │ │ │
│ BRIDGE-TBI │ │ Research-as-a- │ │ RAIGH Academy │
│ Aya Clinician │ │ Service │ │ AI workforce │
│ TeleCare Africa │ │ Evidence │ │ development for │
│ │ │ Capsules │ │ global health │
│ For: Clinicians │ │ For: Researchers│ │ For: Universities│
└─────────────────┘ └─────────────────┘ └─────────────────┘
│
┌────────▼────────┐
│AIDA Infrastructure│
│ │
│ Sovereign data │
│ for governments │
│ │
│ For: Ministries │
│ of Health │
└─────────────────┘
Current focus: TBI (first vertical). The methodology is domain-agnostic — each new clinical domain requires a domain ontology + constraints + domain experts, not new infrastructure. Expansion roadmap: Sepsis → Cardiovascular → Stroke → Surgical Risk.
Community assets for the health AI ecosystem:
| Resource | Description |
|---|---|
| awesome-health-ai-skills | 17 executable health AI skills organized in 7 bundles — the Elements of Health AI |
| awesome-health-ai-regulations | Global regulatory frameworks (FDA, EU AI Act, Africa, Asia-Pacific) with structured JSON data |
| awesome-clinical-prediction-models | Living registry of clinical prediction models with YAML-structured evaluations |
| awesome-ai-evaluation-metrics | Cross-modality evaluation taxonomy for health AI systems |
SDG Alignment: SDG 3 (Good Health — Targets 3.4, 3.8, 3.d) · SDG 4 (Quality Education — Target 4.4) · SDG 9 (Innovation — Target 9.5) · SDG 17 (Partnerships — Target 17.6)
EvidenceOS operates through four legal entities with distinct roles — separating commercial product development from independent evidence production and open-source stewardship:
| Entity | Role | Jurisdiction |
|---|---|---|
| EvidenceOS Inc. | Product company, IP holder | US (Delaware) |
| EvidenceOS OÜ | EU operations, CE Mark pathway | Estonia |
| CEAIH | Independent evidence production institute, grant applicant (NIH, Gates, Wellcome) | US 501(c)(3) |
| PAATHI | Open-source stewardship, LMIC access, continental AI governance | Rwanda (CLG) |
Data policy: This organization contains NO real patient data. All included data is synthetically generated. Real patient data is handled in separate, sovereign environments compliant with HIPAA, GDPR, and national data laws. See DATA_POLICY.md for details.
Licensing: Research packages are Apache-2.0. Awesome-lists are CC-BY-4.0. Platform code follows dual licensing — see individual repo LICENSE files.
|
Researchers & Academics
|
Clinicians & Health Systems
|
|
Governments & Development Partners
|
Builders & Contributors
|
Website · Research Packages · Platform · Contact
Apache-2.0 · Built in Delaware, Turku, Kigali, and everywhere clinical AI needs to prove it works.