Incubator
Community Partner
Accelerator
Carob captures the reasoning behind real medical decisions, transforms it into de-identified structured datasets for AI training, and partners with communities to develop patient-centric benchmarks and educational tools.
Clinical reasoning data must reflect the full diversity of patients, settings, and health backgrounds worldwide. Our business model is revenue share. We believe the institutions and communities generating this data need to be the direct financial beneficiaries.
When expert clinicians work through a difficult case together, they weigh evidence, challenge each other, and change their minds — and none of it is captured. Carob records that deliberation, de-identifies it, and structures it into datasets, benchmarks, and educational tools. We do it in partnership with the institutions and communities who generate the data, on a revenue-share model.
Today's medical AI is trained on textbooks, exams, and case write-ups — the polished end product. The deliberation that produced the diagnosis is discarded, so models imitate conclusions without the reasoning underneath.
Hospitals and communities generate the data that trains commercial AI, and rarely see any of the value. That breeds distrust and keeps the best data locked away.
Model evaluations lean on exam-style questions that miss the diversity of patients, settings, and health backgrounds clinicians actually see.
Structured, de-identified clinical reasoning data for training and evaluating frontier diagnostic models.
Revenue-share data partnerships that turn existing clinical activity — like case conferences — into a durable, ethical data asset.
Patient-centric benchmarks and educational tools built with the communities the data comes from.
De-identified, structured records of real specialist case deliberations — the questions asked, hypotheses raised, and evidence weighed — licensed for AI training.
Evaluation sets developed with clinical communities, designed to test diagnostic reasoning across the diversity of real patients and settings.
Training material built from the same corpus, returning value directly to the teaching institutions that generate it.
Want a walkthrough or a sample of the data schema? Request a demo.
Cognitive science & HCI (Carnegie Mellon). MBA (Imperial College London). Engineering at Morgan Stanley — fixed-income derivatives and data pipelines.
Engineer at Meta (NYC). NLP background from Johns Hopkins. Owns the data and model stack.
Wharton MBA. Private equity and finance/operations leadership. Owns the financial model, go-to-market motion, and capital.
Supported by part-time ML engineering, and clinical & academic advisors across pathology and critical-data research.
Fill out the form and we'll get back to you within one business day.
Follow Carob on LinkedIn