Can Your AI Analyst Keep a Secret? - Aaron Linsky, Bridgewater & Nikhil Benesch, turbopuffer
TurboPuffer was built to enforce per-document access rights in enterprise AI RAG pipelines
“Each email, meeting note, or report can have its own unique set of access rights, belonging to different groups or individual users.”
Bridgewater Associates and TurboPuffer jointly presented the challenge of building AI analysts over private enterprise data where each document carries its own access permissions. Unlike public web data, internal documents require fine-grained, per-user retrieval controls that standard vector databases don't handle natively. TurboPuffer was developed specifically to solve this access-rights problem in AI analytics pipelines, with Bridgewater as a production use case.