AstraZeneca Keeps OLTP and OLAP Separate
AI program manager at AstraZeneca on running self-hosted ClickHouse
The key implication is that ClickHouse is being used as a fast read layer inside a regulated stack, not as the place where AstraZeneca wants its official system of record to live. In practice, patient and clinical source data stays in governed operational systems, while ClickHouse handles sub second retrieval, vector search, and analytics for AI workflows. That separation fits pharma rules around validated systems, auditability, and tight control over how sensitive records are stored and changed.
-
The rest of the interview shows a clean split of labor. Snowflake handles regulatory reporting, complex transformations, and cross domain governance, while ClickHouse is the speed layer for real time analytics and RAG over patient records. That makes the refusal to merge OLTP and OLAP a design principle across the stack, not a one off objection to Postgres integration.
-
ClickHouse is now pushing a unified Postgres plus analytics story, where Postgres remains the transactional system of record and ClickHouse executes analytics through native sync and query extensions. That pitch is attractive for software teams trying to reduce pipeline complexity, but this AstraZeneca workflow shows why regulated pharma buyers may still prefer a harder wall between write systems and analytical systems.
-
This is consistent with pharma compliance norms. FDA guidance treats electronic record systems as validated environments where data integrity, change control, and documented procedures matter. AstraZeneca also states that patient and health data are coded, protected, and overseen under strict privacy and AI controls. Separate systems make those controls easier to prove during review.
The likely direction is more integration at the query layer, but continued separation at the system layer. ClickHouse can keep expanding as the retrieval and inference engine for compliant AI applications, while transactional patient data remains in tightly governed source systems. In regulated enterprises, the winning architecture is not one database for everything, but a fast analytics engine that plugs into a validated record keeping stack without replacing it.