Vitruvia
Vitruvia provides Clinker, an autonomous-agent pipeline that converts unstructured clinical text into structured, interoperable data for analytics and healthcare model training. The platform preserves clinical context such as presence, experiencer, and temporality, and links concepts across ontologies including SNOMED CT, ICD, LOINC, DSM, and OMOP. In an internal MIMIC-III evaluation, Clinker achieved an 89.2% SNOMED-CT linking accuracy and a named entity recognition F1 score of 0.850.
- Healthcare Technology
- Artificial Intelligence
- AI Agents
- Digital Health
- Data & Analytics
- Software Only
Funding
Founders
Founder details are not available yet.
Product
Problem
Around 80% of healthcare data exists as unstructured free text, including clinical notes, reports, and discharge letters. Varying writing styles and inconsistent abbreviations make it difficult to use this information at scale for analytics, research, or model training.
Solution
Vitruvia offers Clinker, a pipeline of autonomous agents that structure clinical text with traceable reasoning at every step. The system ingests, de-identifies, and tokenizes clinical narratives, then uses specialist agents to extract clinical concepts while preserving context such as presence, experiencer, and temporality. Clinker links these concepts across major ontologies including SNOMED CT, ICD, LOINC, DSM, and OMOP, producing interoperable data suitable for analytics, risk prediction, simulation, and healthcare model training. The platform is designed for privacy-preserving, sovereign deployments.
Target Audience
Primary customers include healthcare systems, payers, life sciences researchers, and healthcare AI developers who need to convert unstructured clinical narratives into structured, standards-compliant data for analytics and model training.
Features
- Specialist autonomous agents that identify clinical concepts and preserve sentinel context including presence, experiencer, and temporality
- Ontology-flexible concept linking across SNOMED CT, ICD, LOINC, DSM, and OMOP
- Coverage spanning approximately 500,000 clinical entities
- Achieved 89.2% SNOMED-CT linking accuracy in an internal MIMIC-III evaluation, with a named entity recognition F1 score of 0.850
- Auditable pipeline with traceable reasoning for regulatory and clinical confidence
- Built for privacy-preserving, sovereign deployments suitable for sensitive healthcare environments