Base de Busca Jurídica

Search foundations are implemented at the data layer

Leci already includes structural prerequisites for legal search, even though full API/product search flows are still roadmap work. This allows incremental implementation without redesigning the core schema.

Implemented search primitives

Current code provides:

generated Portuguese tsvector on document_nodes;
GIN index for full-text search performance;
embeddings table with vector(768) column;
IVFFlat vector index for similarity retrieval.

Practical implication today

The database is ready for full-text and semantic query strategies, but no internal API route currently exposes a production search contract.

🚧 Planned Feature — Endpoint-level search contracts, ranking strategies, and pagination semantics are part of roadmap milestones.

Relevance and quality direction

Planned relevance model should include:

lexical precision baseline (FTS);
optional semantic reranking where data quality supports it;
explicit traceability of match origin for legal confidence.

Performance considerations

Search performance risk grows with ingestion scale and combined lexical/vector querying.