Vector database
A storage system that indexes embeddings to find very quickly the items closest to a query, the central infrastructure of assistants built on internal documentation, and a new exposure surface for data.
A vector database stores embeddings, along with the texts or documents they represent, and can find within milliseconds, among millions of items, those whose vector is closest to a query vector. It relies on approximate nearest neighbor search algorithms, which trade a little accuracy for considerable speed. It has become the central infrastructure of retrieval-augmented generation systems: when a staff member asks an assistant about underwriting policy or a contract, the question is turned into a vector, the database retrieves the relevant passages of documentation, and the language model writes its answer from them. For the insurer, two points call for vigilance. The quality of answers depends on what the database contains: outdated documents, conflicting versions and unverified content end up in the answers. And the database concentrates a copy of often sensitive documents, contracts, customer data, adjusters' reports, in a form whose access rights management is less mature than that of conventional systems, making it a new target and a new data protection compliance issue.
Textbook case: an insurer indexes all its claims files in a vector database for its internal assistant. An audit reveals that the assistant quotes excerpts from medical records to staff who have no access to them in the claims system, because permissions were not replicated in the index: the tool is suspended until results are filtered by authorization.
vector database, vector store, vector index
Follow the research
One letter a week. First, the week's underwriting note, when one is published: a dated, argued decision, signed by Paul Bernardski. Then what the observatory published on cyber risk, political risk and insurance. No account, no tracking.
No spam, one-click unsubscribe, no data ever sold.