Struct TieredEngine 

pub struct TieredEngine { /* private fields */ }

Tiered embedding engine: fast writes via static backend, quality reads via transformer.

Enabled with DAKERA_TIERED=1. When disabled, embed_for_write and embed_for_read both delegate to the single quality backend (no tier split).

Implementations

impl TieredEngine

pub async fn new(config: &ModelConfig) -> Result<Self>

Build from the given ModelConfig.

When DAKERA_TIERED=1, initialises both the static fast backend and the transformer quality backend. Otherwise both slots use the quality backend.

pub async fn embed_for_write(&self, texts: &[String]) -> Result<Vec<Vec<f32>>>

Embed texts for storage (write path).

Uses the fast backend when tiered mode is enabled, quality backend otherwise. Returns embeddings suitable for indexing — may be static Model2Vec quality.

pub async fn embed_for_read(&self, texts: &[String]) -> Result<Vec<Vec<f32>>>

Embed texts for retrieval scoring (read path).

Always uses the quality backend so recall scoring is at transformer quality.

pub async fn embed_query(&self, query: &str) -> Result<Vec<f32>>

Embed a single query for retrieval.

pub fn is_tiered(&self) -> bool

Whether the tiered split is active (fast writes / quality reads).

pub fn fast_dimension(&self) -> usize

Dimension of the fast (write) backend.

pub fn quality_dimension(&self) -> usize

Dimension of the quality (read) backend.

pub fn fast_backend(&self) -> Arc<dyn EmbeddingBackend>

Expose fast backend for background re-embedding tasks.

pub fn quality_backend(&self) -> Arc<dyn EmbeddingBackend>

Expose quality backend for background re-embedding tasks.