All Lessons

Under what operational condition should an engineering team choose an exact Flat (brute-force kNN) vector index over an Approximate Nearest Neighbor (HNSW) index?

1 / 1
intermediate 45 min

Day 9: Build a Vector Index and Understand Its Tradeoffs

Build a Vector Index and Understand Its Tradeoffs

Loading lesson content...

In the HNSW index configuration above, what is the direct operational trade-off of increasing ef_search from 64 to 256 at query time?

1 / 1

PolicyDesk stores 200,000 policy chunks across 4 jurisdictions in an HNSW vector index. When German employees query parental leave policies (jurisdiction == 'DE', representing 2 percent of total chunks), naive post-filtering drops 98 percent of retrieved neighbors, resulting in recall collapse. Which index-level mechanism resolves this?

1 / 1