Tuning HNSW parameters for filtered search

Hosted by Radu Gheorghe and Doug Turnbull

Thu, Apr 16, 2026

5:00 PM UTC (1 hour)

Virtual (Zoom)

Free to join

Invite your network

Go deeper with a course

Cheat at Search with Agents
Doug Turnbull
View syllabus

What you'll learn

Why filters make HNSW vector search slow

We'll start with a short intro on how HNSW works and why filters are expensive and hurt recall.

Parameters to improve performance/recall

Pros and cons of post filters, ACORN-1, brute force kNN, overfetching, and adaptive beam search.

Tools to tune vector search filter search

How HNSWTuner+VespaNNParameterOptimizer allows you to change these knobs and see the impact of latency and recall

Why this topic matters

Most vector search involves filtering, which has a big impact on both performance and quality compared to unfiltered Approximate Nearest Neighbor search on an HNSW. We'll discuss ways to limit this impact, keeping queries nice and fast.

You'll learn from

Radu Gheorghe

Software Engineer, Vespa.ai

Radu has been in the search space for many years, mainly on Elasticsearch, Solr, OpenSearch, and, more recently, Vespa.ai. Helps users with both the relevance and the operations side of retrieval. Enjoys education in all its forms (training, blog posts, books, conferences...) and got the chance to be involved in all of them.


Doug Turnbull

Retrieval + Snuggie Enthusiast

In 2012, Doug got bit by the search bug and he's still trying to keep up. From full-text search, to Learning to Rank models, to search agents that generate their own code, he knows the overwhelming landscape first hand. Yet Doug still works to deeply understand the what / how / why, and help teams use these technologies practically, distinguishing hype from reality.

He’s led search at Reddit, Shopify, and Wikipedia, authored Relevant Search and AI Powered Search, and advised 100+ organizations over the years - all in pursuit of the same question: how does search actually work?

Previously at

Vespa.Ai
Reddit
Shopify.com
Wikipedia

Sign up to join this lesson

By continuing, you agree to Maven's Terms and Privacy Policy.