Constructive Search

Constructive Search

All search strategies available on the Constructive platform — from keyword search to semantic vector similarity, unified into a single composable system.

When to Apply

Use this skill when:

• Adding any kind of search to a table (text, semantic, spatial)
• Choosing between search strategies for a use case
• Using the unified unifiedSearch / searchScore system
• Configuring Search* blueprint nodes (SearchUnified, SearchVector, SearchFullText)
• Working with the graphile-search plugin and its adapters
• Building RAG (retrieval-augmented generation) pipelines

Search Strategy Overview

Quick Start: Unified Search

The simplest way — unifiedSearch fans a single string to all text-compatible algorithms:

const results = await db.article.findMany({
  where: { unifiedSearch: 'machine learning' },
  orderBy: 'SEARCH_SCORE_DESC',
  select: { title: true, searchScore: true },
}).execute();

Blueprint Nodes

SearchUnified (full stack — recommended for most tables)

Orchestrates embedding + BM25 + optional FTS + optional trigram:

{ "$type": "SearchUnified", "data": {
  "embedding": { "source_fields": ["name", "description"] },
  "bm25": { "field_name": "embedding_text" },
  "full_text_search": {
    "field_name": "search_tsv",
    "source_fields": [
      { "field": "name", "weight": "A" },
      { "field": "description", "weight": "B" }
    ]
  },
  "trgm_fields": ["name"]
}}

SearchVector (standalone embeddings)

For tables needing only vector search (no BM25/FTS/trigram):

{ "$type": "SearchVector", "data": { "field_name": "embedding" } }

SearchFullText (standalone tsvector)

For tables needing only full-text search:

{ "$type": "SearchFullText", "data": {
  "field_name": "search_tsv",
  "source_fields": [{ "field": "title", "weight": "A" }]
}}

graphile-search Plugin

Unified PostGraphile v5 search plugin that consolidates all strategies into a single adapter-based architecture. Each strategy is a SearchAdapter:

• TsvectorAdapter — PostgreSQL full-text search
• Bm25Adapter — pg_textsearch BM25 ranking
• TrgmAdapter — pg_trgm fuzzy matching
• PgvectorAdapter — HNSW vector similarity
• PostgisAdapter — spatial distance queries

References

Cross-References

• AI and embedding pipeline: constructive-agents
• Blueprint nodes: constructive-blueprints
• ORM query patterns: constructive-orm
• i18n multilingual search: constructive-i18n