OntologerMed Models

Classifies what the user is asking for — commercialization plan, pathway routing, predicate search, etc.

Checks whether the user provided enough device detail to give a useful answer.

Determines which FDA regulatory pathway this device likely needs: 510(k), PMA, or De Novo.

Routes to one of 20 FDA medical specialty advisory panels.

Maps to the applicable Code of Federal Regulations part (Title 21).

Identifies which FDA product code family this device belongs to (23 families).

Runs all 6 classifiers on the same input text: intent → sufficiency → pathway → panel → CFR → product code.

Population-shaped 256-dim embedding from eligibility criteria. Trials enrolling similar patients cluster together.

Mechanism-of-action 256-dim embedding. Trials targeting the same biological pathway cluster together.

Outcome-shaped 256-dim embedding. Trials with similar success/failure patterns cluster together.

Calibrated probability of clinical trial success. Combines PACT + MOAt + FATE into a 768-dim vector → gradient-boosted ensemble + Platt calibration.

Embeds trial text through PACT + MOAt + FATE, concatenates to 768-dim, runs ORACLE ensemble for calibrated success probability.

Resolves NPI-identified physicians to OpenAlex academic author profiles using multi-tower entity resolution.

Domain-adapted 0.8B LM trained on 551K ClinicalTrials.gov records. Extraction, summarisation, PICO, phase classification, outcome scoring.

Domain-adapted 0.8B LM for commercial intelligence, market analysis, and healthcare commercialization.