vector-index-tuning

Optimize vector index performance for latency, recall, and memory. Use when tuning HNSW parameters, selecting quantization strategies, or scaling vector search infrastructure.

1.27x

Quality

71%

Does it follow best practices?

Impact

84%

1.27x

Average score across 3 eval scenarios

Securityby

Passed

No known issues

Optimize this skill with Tessl

npx tessl skill review --optimize ./plugins/llm-application-dev/skills/vector-index-tuning/SKILL.md

Evaluation results

74%

-4%

HNSW Index Tuning Analysis for Semantic Product Search

HNSW parameter benchmarking

Criteria

Without context

With context

hnswlib used

100%

Cosine space

100%

Recall via set intersection

100%

Memory formula

60%

100%

M recommendation for 500K

100%

ef_construction for 500K

100%

ef_search for 95% recall

Results artifact produced

100%

Benchmark sweeps M and ef_construction

100%

Benchmark sweeps ef_search

100%

Recommendation provided

100%

Search time in ms

100%

83%

31%

Qdrant Collection Setup for Enterprise Document Retrieval

Qdrant collection configuration

Criteria

Without context

With context

qdrant_client used

75%

100%

COSINE distance

100%

Recall HNSW config

50%

100%

Recall: no quantization

100%

Speed HNSW config

25%

Speed: INT8 always_ram=True

77%

Memory HNSW config

42%

100%

Memory: ProductQuantization X16

100%

Balanced HNSW config

44%

100%

Optimizer thresholds

37%

100%

Search params: 0.99 threshold

25%

100%

Search params: 0.95 threshold

44%

100%

95%

25%

Memory Planning for Multi-Scale Vector Search Infrastructure

Vector quantization and memory estimation

Criteria

Without context

With context

FP32 bytes=4

100%

FP16 bytes=2

100%

INT8 bytes=1

100%

PQ bytes~0.05

37%

100%

Binary bytes=0.125

100%

HNSW overhead formula

70%

100%

IVF overhead formula

20%

100%

Index recommendation: <10K → Flat

100%

Index recommendation: 10K-1M → HNSW

100%

Index recommendation: >100M → IVF+PQ/DiskANN

40%

80%

Index recommendation: 1M-100M → HNSW+Quantization

62%

memory_report.json produced

100%

M parameter in HNSW overhead

100%

Repository: wshobson/agents
Commit: 91fe43e

Evaluated: 27 days ago
Agent: Claude Code
Model: Claude Sonnet 4.6

Table of Contents

HNSW Index Tuning Analysis for Semantic Product Search Qdrant Collection Setup for Enterprise Document Retrieval Memory Planning for Multi-Scale Vector Search Infrastructure

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.