Episode

The Patient is Not a Document: Foundation Models for Biomedical AI with Standard BioModel

Podcast
Data in Biotech
Published
Apr 15, 2026
Duration seconds
2994
Processing state
not_requested
Canonical source
https://www.corrdyn.com/
Audio
https://downloads.pod.co/5fac40cd-043b-485f-838d-c63fa01ae1b2/b81bfb9d-8f95-4b2d-a458-6685726b9d55.mp3
JSON
/v1/public/podcasts/data-in-biotech-6628050/episodes/the-patient-is-not-a-document-foundation-models-for-biomedical-ai-with-standard-biomodel
Markdown
/podcast/data-in-biotech-6628050/the-patient-is-not-a-document-foundation-models-for-biomedical-ai-with-standard-biomodel.md

Actions

  • POST https://stenobird.com/v1/public/podcasts/data-in-biotech-6628050/episodes/the-patient-is-not-a-document-foundation-models-for-biomedical-ai-with-standard-biomodel/transcription-requests
    Idempotently request low-priority transcript generation for this episode.
  • GET https://stenobird.com/podcast/data-in-biotech-6628050/the-patient-is-not-a-document-foundation-models-for-biomedical-ai-with-standard-biomodel.md
    Read the agent-friendly Markdown representation of this episode resource.

Summary

In this episode of Data in Biotech, host Ross Katz sits down with Kevin Brown, co-founder of Standard BioModel, to explore one of the most ambitious projects in biomedical AI, building a multimodal foundation model that represents the full complexity of a patient across time. Drawing on a career spanning brain-computer interfaces, computer-aided diagnosis at Siemens Healthineers, and oncology data science at Bristol Myers Squibb, Kevin shares the scientific and philosophical journey that led him to a single conviction: a patient is not a document. Rather than reducing a patient to clinical notes, ICD-10 codes, or isolated test results, Standard BioModel's approach maps every available modality - CT imaging, digital pathology, genomics, EKGs, longitudinal EHR data - into a shared latent space, and models how that patient moves through time. The result is a framework designed not just for prediction, but for counterfactual reasoning, clinical trial matching, and personalized intervention, with open-source models already being validated across leading academic medical centers. What you’ll learn in this episode: >> Why reducing a patient to text - clinical notes, radiology reports, genomic assay summaries - and how mapping multimodal data into a shared latent embedding space preserves information that never makes it into the written record >> How Standard BioModel's temporal architecture models patients as trajectories through an abstract embedding space rather than static snapshots, enabling counterfactual reasoning about the likely impact of interventions on a patient's future health trajectory >> Why no single foundation model can own every clinical vertical and how building a highly generalizable base model that facilitates downstream fine-tuning is a…