Clinical discovery
Refining the target workflow with clinical and institutional stakeholders.
Technologies CelestyxAI Inc.
Building clinical AI infrastructure that institutions can inspect, govern and validate.
CelestyxAI is a Canadian health technology company focused on professional-facing clinical intake and pre-assessment support. The company is developing a safety-oriented architecture that combines explicit policy controls, bounded AI assistance, structured outputs and human validation.
Company focus
Clinical workflow support · AI safety architecture · structured intake · auditability · interoperability · evidence-driven validation
Mission
Make AI useful in clinical intake without making it opaque or autonomous.
Healthcare organizations need more than a capable model. They need workflow fit, clear accountability, testable safeguards, traceability and a path to integration. CelestyxAI is being built around those institutional requirements from the outset.
Professional controlThe healthcare professional remains responsible for review and clinical decision-making.
Explicit safeguardsSafety-critical rules should be represented and versioned outside the LLM.
Evidence before claimsValidation milestones must precede clinical outcome claims.
Institutional readinessPrivacy, governance, auditability and interoperability are product requirements, not later add-ons.
Current priorities
What we are working on now.
Safety validation
Defining test cases, failure modes, escalation logic and evaluation criteria.
Target architecture
Separating deterministic policy logic, model orchestration, structured validation and audit traces.
Design partnerships
Building relationships with institutions, researchers and digital-health ecosystems that can support a controlled pilot pathway.
How we work
Designed to evolve with evidence.
Discover
Understand the real workflow
Map users, handoffs, data, failure points, escalation paths and institutional constraints before choosing automation.
Constrain
Define where AI is allowed to help
Separate deterministic controls from generative tasks and specify what the model may and may not do.
Validate
Test behaviour before deployment
Use expert review, adversarial cases, workflow testing and measurable endpoints.
Monitor
Treat deployment as the start of evaluation
Version changes, incident review and performance drift should remain visible after launch.
Collaborate