Summary

Before State: Customers would benchmark using third party libraries, use HNSW even for small use cases, and have to use workaround strategies for multi-tenancy.

Pains: Unclear to a customer how to measure recall of their systems, extra resource overhead for small vector use cases, challenging to provide multi tenanted use cases

After State: Exact search can be executed directly within the $vectorSearch stage, unlocking a set of use cases.

Impact: More use cases graduate from POC into production as users can evaluate their retrieval accuracy, more vector search use cases unblocked via the same query syntax

Motivation

Who is the affected end user?

Developers looking to build retrieval augmented generation, semantic search, or recommendation systems with vector search who want to benchmark their systems, are working with a small scale of vectors, or have multi-tenant use cases.

These customers can be both small startups and large companies.

Documentation

Product Description

Scope

Tech Design

Syntax