This series covers how Snowflake® and VaultSpeed work together, one industry per post: the business problem, the architecture, and what changes once it is in place. Part 1 covered financial services. Healthcare and life sciences share that problem, with two complications: the data is far more personal, and the source systems rarely agree on who the patient is.
In short: Snowflake gives clinicians, analysts and agents governed access to healthcare data, and VaultSpeed keeps the patient-matching logic, lineage and code beneath it correct. VaultSpeed's agents propose how source records map to shared concepts such as Patient and Encounter, with evidence behind every match. Snowflake semantic views and Cortex Agents answer on top, masked by role. When a new study site, protocol amendment or payer feed arrives, the update lands as one reviewed pull request, with lineage from every reported figure back to the source system.
Three use cases show the pattern. Clinical trials pull from electronic data capture, labs, devices and research partners, and every new study or protocol amendment adds mappings that are usually maintained by hand. A patient or member 360 has to combine the EHR, claims, care management and partner feeds, each identifying the same person differently. In the revenue cycle, a denied claim usually traces back to two systems describing the same encounter in different ways.
All three share one gap: the data model and the code that integrates the data were built separately, and they drifted apart. When a new source arrives or a reporting rule moves, the work starts with archaeology.
VaultSpeed is an agentic data modeling platform: bring AI to data modeling and automate the code. Its agents take on the matching work that usually falls to analysts with spreadsheets. They profile the EHR, claims, trial and care management sources, compare identifiers, demographics and encounter dates, and propose how each record maps to shared concepts such as Patient, Member, Provider, Encounter, Claim, Study and Lab Result.
Every proposed match rule comes with the evidence behind it, ambiguous cases are flagged for a data steward, and people and agents review each other's work before anything ships. The context store holds the data model, lineage and data products in one governed graph, and generators write the integration code from it as runtime-free SQL or dbt in the organization's own Git. Where a Data Vault fits, it keeps the full history of every value and correction, so an amended lab result or an auditor's question can always be traced.
Snowflake is where that code runs and where clinicians and analysts ask their questions. Semantic views define measures such as 30-day readmission rate and denial rate by payer once. Cortex Agents combine Cortex Analyst over those views with Cortex Search over discharge summaries and trial protocols, so a care coordinator in Snowflake CoWork (formerly Snowflake Intelligence) can ask why a patient's readmission risk rose and get an answer that joins claims history to the latest discharge note. Snowflake Horizon Catalog masking policies keep protected health information visible only to the roles that need it.
Two things make that acceptable. The first is containment. VaultSpeed runs its data plane in the organization's own cloud and works from metadata, so patient data stays within the organization's environment. Every agent run executes in its own sandbox, created for that task and removed when it ends, and agents hold no standing credentials to the warehouse: the platform grants scoped access for the length of the run and no longer. The organization caps agent spend and chooses which LLM its agents use.
The second is change. When a new study site, protocol amendment or payer feed arrives, agents harvest the new metadata and propose the updated mappings, the change report shows every affected object, and the regenerated code arrives as a reviewed pull request.
That split holds for the rest of the series: Snowflake gives people and agents governed access to the data, and VaultSpeed keeps the data model, lineage and code beneath it correct. Manufacturing is next, where the data comes from machines as well as records.
VaultSpeed is available as a Snowflake Native App on Snowflake Marketplace, so you can install it in your own Snowflake account and start the 30-day free trial straight from the listing. To see how the agents work before you install, take a look at the platform or the VaultSpeed and Snowflake page.
Snowflake gives clinicians, analysts and agents governed access to the data, and VaultSpeed keeps the patient-matching logic, lineage and code beneath it correct. VaultSpeed's agents propose how source records map to shared concepts such as Patient and Encounter, and Snowflake semantic views and Cortex Agents answer on top.
Agents profile each source, compare identifiers, demographics and encounter dates, and propose how each record maps to shared concepts such as Patient, Member, Provider, Encounter, Claim, Study and Lab Result. Every proposed match comes with the evidence behind it, and ambiguous cases are flagged for a data steward to review.
VaultSpeed runs its data plane in the organization's own cloud and works from metadata, so patient data stays within the organization's environment. Every agent run executes in its own sandbox with scoped access for the length of the run only, and the organization caps agent spend and chooses which LLM its agents use.
Agents harvest the new metadata and propose the updated mappings, the change report shows every affected object, and the regenerated code arrives as a reviewed pull request. Once it merges, Snowflake's agents pick up the change the same day.
VaultSpeed is available as a Snowflake Native App on Snowflake Marketplace. You can install it in your own Snowflake account and start the 30-day free trial straight from the listing.