Build and maintain retrieval indexes

Derive retrieval index behavior from metric choices, graph or partition structure, compression, filters, update paths, and measured recall.

10 published lessons1 module240 min of guided study
  1. 01
    Build and maintain retrieval indexesadvanced · 10 lessons

Follow the path or begin with a question. Every lesson is open.

Your notebook0 of 10 lessons completed

Your place, at your pace. Progress is optional and saved only in this browser when available.

Inside this module

Each lesson includes an interactive lab

Build and maintain retrieval indexes

Derive retrieval index behavior from metric choices, graph or partition structure, compression, filters, update paths, and measured recall.

  1. 01

    Choose the geometry before choosing the index

    Distinguish cosine, inner product, and squared Euclidean ranking, including normalization and zero-vector policy.

    24 min
  2. 02

    Probe partitions without confusing them with answers

    Derive IVF candidate selection, nprobe cost, boundary misses, and the conditions for exhaustive IVFFlat search.

    24 min
  3. 03

    Compute approximate distances from compact codes

    Derive product-code storage, asymmetric distance tables, and the distinction between reconstruction and ranking error.

    24 min
  4. 04

    Search a graph with a bounded candidate frontier

    Separate HNSW hierarchy, base-layer exploration, result count, and implementation-specific search parameters.

    24 min
  5. 05

    Build useful routes instead of only short edges

    Explain neighbor diversity, construction search, degree accounting, and why graph quality limits later queries.

    24 min
  6. 06

    Search the eligible population without losing its routes

    Define filtered ground truth and compare postfiltering, traversal restrictions, and selective exact fallback.

    24 min
  7. 07

    Keep logical documents correct while index bytes change

    Separate stable document identity, logical deletion, physical removal, delta overlays, and rebuild cutovers.

    24 min
  8. 08

    Merge shard answers under an explicit exactness contract

    Prove when local top-k suffices, then identify approximation, replica, and deduplication cases that break the proof.

    24 min
  9. 09

    Measure a frontier instead of selecting one flattering number

    Define exact recall with ties, held-out queries, bounded uncertainty, and comparable latency-memory operating points.

    24 min
  10. 10

    Promote a compatible retrieval system as one immutable snapshot

    Bind encoder geometry, codebooks, corpus versions, metadata, and serving parameters into a reviewable artifact.

    24 min
All engineering modules