Similarity Search Patterns

Patterns for implementing efficient similarity search in production systems.

When to Use This Skill

• Building semantic search systems
• Implementing RAG retrieval
• Creating recommendation engines
• Optimizing search latency
• Scaling to millions of vectors
• Combining semantic and keyword search

Core Concepts

1. Distance Metrics

| Metric | Formula | Best For | | —————— | —————— | ——————— | — | ————-- | | Cosine | 1 - (A·B)/(‖A‖‖B‖) | Normalized embeddings | | Euclidean (L2) | √Σ(a-b)² | Raw embeddings | | Dot Product | A·B | Magnitude matters | | Manhattan (L1) | Σ | a-b | | Sparse vectors |

2. Index Types

┌─────────────────────────────────────────────────┐
│                 Index Types                      │
├─────────────┬───────────────┬───────────────────┤
│    Flat     │     HNSW      │    IVF+PQ         │
│ (Exact)     │ (Graph-based) │ (Quantized)       │
├─────────────┼───────────────┼───────────────────┤
│ O(n) search │ O(log n)      │ O(√n)             │
│ 100% recall │ ~95-99%       │ ~90-95%           │
│ Small data  │ Medium-Large  │ Very Large        │
└─────────────┴───────────────┴───────────────────┘

Templates and detailed worked examples

Full template library and detailed worked examples live in references/details.md. Read that file when you need the concrete templates.

Best Practices

Do’s

• Use appropriate index - HNSW for most cases
• Tune parameters - ef_search, nprobe for recall/speed
• Implement hybrid search - Combine with keyword search
• Monitor recall - Measure search quality
• Pre-filter when possible - Reduce search space

Don’ts

• Don’t skip evaluation - Measure before optimizing
• Don’t over-index - Start with flat, scale up
• Don’t ignore latency - P99 matters for UX
• Don’t forget costs - Vector storage adds up