Faster phrase search with shingled bloom filters in Brainstore - Blog - Braintrust

How trigram shingles in Brainstore's bloom filters keep phrase search fast as trace data grows large and cold.

7 July 2026Ankur Goyal7 min When you debug an agent, you often start from a bit of text you saw somewhere, maybe a line in a Slack screenshot, a snippet from an X post, or an output you want to trace back through your logs. Say you search for "They also touched on AI provider modifications." Every word in it shows up in almost every trace, but the exact wording, in that order, is rare. Common terms with a rare intersection is the worst case for a traditional search system, and at agent scale, where a dataset can balloon past 100TB of text, it is the difference between an instant answer and…

saved by

• Benedict Neo

related reading

• From grep to SPLADE: A Journey Through Semantic Search - Elicitblog.elicit.com
• Fast regex search: indexing text for agent tools · Cursorcursor.com
• Building a web search engine from scratch in two months with 3 billion neural embeddingsblog.wilsonl.in
• The vector database to build knowledgeable AI | Pineconepinecone.io
• Optimizing Postgres Text Search with Trigrams - Alex Klibiszalexklibisz.com
• Regular Expression Matching with a Trigram Indexswtch.com
• turbopufferturbopuffer.com
• Faiss: A library for efficient similarity search - Engineering at Metaengineering.fb.com
• Chroma Context-1: Training a Self-Editing Search Agent | Chromatrychroma.com
• GitHub - thesephist/monocle: Universal personal search engine, powered by a full text search algorithm written in pure Ink, indexing Linus's blogs and private note archives, contacts, tweets, and over a decade of journals.github.com
• Bloom Filters - Much, much more than a space efficient hashmap! | Ben E. C. Boyterboyter.org
• Create an advanced search engine with PostgreSQL | xataxata.io