platform-search

Platform-wide hybrid search: pgvector + FTS with Reciprocal Rank Fusion (ADR-087)

Overview

Combines PostgreSQL pgvector semantic search with full-text search (FTS), merged via Reciprocal Rank Fusion (RRF). Optional MMR diversity filter and temporal decay scoring.

Installation

pip install -e packages/platform-search

Usage

from platform_search.service import SearchService

results = SearchService.search(
    query="Gefährdungsbeurteilung Arbeitsplatz",
    tenant_id="abc-123",
    source_types=["assessment"],
    top_k=10,
)

Configuration

# config/settings/base.py
PLATFORM_SEARCH = {
    "EMBEDDING_PROVIDER": "openai",
    "EMBEDDING_MODEL": "text-embedding-3-small",
    "EMBEDDING_DIMENSIONS": 1536,
    "VECTOR_WEIGHT": 0.6,
    "TEXT_WEIGHT": 0.4,
    "RRF_K": 60,
    "DEFAULT_TOP_K": 10,
}

Requires OPENAI_API_KEY in Django settings (ADR-045).

Database

Migration creates search_chunks table in content_store DB (ADR-062). Run: python manage.py migrate platform_search --database=content_store

Related ADRs

• ADR-087: Hybrid Search Architecture
• ADR-062: Content Store
• ADR-045: Secret Management
• ADR-072: Schema Isolation (Row-Level deviation documented)