Portiere is a medical data mapping platform that converts clinical source data into OMOP CDM and FHIR R4 standards using AI-assisted schema and concept mapping. The platform combines hybrid search with cross-encoder reranking to auto-accept high-confidence mappings, routing 51.8% of codes directly to ETL workflows while flagging the rest for human review. It provides a local review interface and CSV round-trip for clinical subject matter experts.
Funding
Funding not disclosed
Founders
Product
Problem
Healthcare organizations migrating to OMOP or FHIR standards face the labor-intensive task of matching thousands of local codes to standard concepts one by one. The de-facto tool, OHDSI Usagi, relies on TF-IDF string similarity and leaves every mapping decision to a human reviewer, creating months of repetitive work with no confidence signals or audit trails. Each institution repeats the same mapping effort independently, locking results in spreadsheets with no way to reuse the work.
Solution
Portiere automates the safe portion of medical vocabulary mapping while queuing uncertain cases for human review. The platform retrieves candidate concepts using hybrid lexical and biomedical-embedding search, then reranks them with a cross-encoder model. Every code is routed by confidence score: high-confidence mappings skip review entirely and flow directly into ETL pipelines, while lower-confidence codes present reviewers with ranked candidates rather than a blank search box. The system publishes benchmark results showing 51.8% of codes auto-accept at default thresholds, with the correct concept appearing in the top 10 for 58.8% of queries. Reviewers work through a local UI or CSV round-trip, and administrators can tune routing thresholds to match their risk tolerance.
Target Audience
Primary customers are data engineers, terminologists, and clinical informatics teams at healthcare organizations, research networks, and life sciences companies adopting OMOP CDM or FHIR R4 standards.
Features
- Hybrid retrieval combining lexical search with biomedical embeddings, reranked by a cross-encoder model
- Confidence-based routing with configurable auto-accept and review thresholds
- Benchmark-validated performance: 51.8% auto-accept rate and 58.8% top-10 accuracy on 1,000 ICD-10-CM to SNOMED mappings
- Local review interface with ranked candidate lists for human-in-the-loop validation
- CSV round-trip workflow enabling clinical subject matter experts to review mappings offline
- Audit trail and confidence signals for every mapping decision, supporting compliance and reuse