backend-search-patterns

Backend Search Patterns

Purpose

Design search architecture with indexing strategy, query design, and relevance tuning.

Agent Protocol

Trigger

Exact user phrases: "search", "Elasticsearch", "Meilisearch", "Algolia", "full-text search", "search index", "search query", "faceted search", "autocomplete", "search ranking", "search relevance", "indexing strategy", "search aggregation", "synonym search", "fuzzy search".

Input Context

Before activating, verify:

• Data volume (documents count, average document size, growth rate)
• Search requirements (full-text, faceted navigation, geo-spatial, autocomplete)
• Update frequency (real-time CDC, hourly batch, daily reindex)
• Consistency requirements (eventual consistency acceptable, or need read-your-writes)

Output Artifact

Search architecture design as formatted text.

Response Format

# Index mapping with analyzers
# Indexing strategy (CDC/batch/webhook)

# Query DSL template
# Aggregation patterns

# Cluster config and resource limits

No preamble. No postamble. No explanations. No filler/hedging/transitions. Compress output — why use many token when few do trick.

Completion Criteria

• Search provider selected based on requirements
• Index mapping defined with field types, analyzers, and doc values
• Indexing strategy chosen (CDC/batch/webhook) with sync mechanism
• Search query patterns designed (full-text, faceted, autocomplete, geo)
• Relevance tuning configured (BM25, field boosting, function scoring)
• Operational concerns addressed (aliases, shards, snapshots, monitoring)

Max Response Length

300 lines of mapping, queries, and configuration.

Decision Tree

Which Search Engine?

What kind of search do you need?
  ├── Complex search: aggregations, geo, custom scoring, multi-language
  │   └── Elasticsearch / OpenSearch — full control, steep learning curve
  ├── Simple typo-tolerant full-text search, instant setup
  │   └── Meilisearch — excellent out-of-box relevance, minimal config
  ├── Managed, no ops, global edge network, per-query pricing
  │   └── Algolia — fastest time-to-value for e-commerce and content
  ├── Fast, simpler alternative to Elasticsearch, lower resource usage
  │   └── Typesense — REST-first, good relevance, low memory footprint
  └── Already using AWS ecosystem, simple search needs
      └── OpenSearch Service — managed, familiar if coming from ES 6.x