Can Your AI Analyst Keep a Secret? – Aaron Linsky, Bridgewater & Nikhil Benesch, turbopuffer
Episode details
Episode description
An AI analyst at a firm like Bridgewater needs private documents, but every email, memo and research note carries its own permissions. Aaron Linsky (head of product engineering, Bridgewater Associates) explains why the Vault's 100 million permissioned documents broke their first search setup, and Nikhil Benesch (CTO, turbopuffer) breaks down the fix: post-filtering loses recall, pre-filtering is too slow, and "native filtering" over a clustered vector index gets both, delivering about 120ms P90 searches across 400 million vectors. Chapters 00:00 Intro: why permissions matter for AI analysts 02:07 Bridgewater's research process 03:43 The Vault: 100M documents, many permissions 05:14 Modeling permissions, and where it broke 06:25 What turbopuffer is 08:01 Napkin math: why brute force fails at scale 09:43 How vector indexes work 11:50 Permissions break the index 12:19 Post-filtering vs pre-filtering 14:09 Native filtering: the best of both 15:23 Results: 120ms across 100M documents 15:47 Writing permission filters in practice 17:00 Beyond permissions: isolation-native search Aaron Linsky, Bridgewater Associates Nikhil Benesch, turbopuffer Join us in SF for the second annual AIE CODE summit: https://aie.codes our last flagship summit of 2026. Applications live now!
Watch also from the same channel
More episodes from this channel
You might be interested in
