Struct RelevanceEngine 

pub struct RelevanceEngine { /* private fields */ }

Proactive relevance engine for memory surfacing

Implementations

impl RelevanceEngine

pub fn new(ner: Arc<NeuralNer>) -> Self

Create a new relevance engine

pub fn set_active_ab_test(&self, test_id: Option<String>)

Set active A/B test for this engine

When set, the engine will use weights from the A/B test based on user assignment.

pub fn get_active_ab_test(&self) -> Option<String>

Get active A/B test ID

pub fn get_weights(&self) -> LearnedWeights

Get current learned weights

pub fn set_weights(&self, weights: LearnedWeights)

Set learned weights (e.g., loaded from storage)

pub fn apply_feedback( &self, semantic_contributed: bool, entity_contributed: bool, tag_contributed: bool, helpful: bool, )

Apply feedback to update learned weights

Call this when user indicates a surfaced memory was helpful or not.

pub fn surface_relevant( &self, context: &str, memory_system: &MemorySystem, graph_memory: Option<&GraphMemory>, config: &RelevanceConfig, feedback_store: Option<&RwLock<FeedbackStore>>, ) -> Result<RelevanceResponse>

Surface relevant memories for the given context

This is the main entry point for proactive memory surfacing. Target latency: <30ms

pub fn surface_relevant_with_momentum( &self, context: &str, memory_system: &MemorySystem, graph_memory: Option<&GraphMemory>, config: &RelevanceConfig, momentum_lookup: &HashMap<Uuid, f32>, ) -> Result<RelevanceResponse>

Surface relevant memories with feedback momentum integration (FBK-5)

This variant incorporates momentum EMA from past feedback into scoring. Memories with positive momentum (consistently helpful) get boosted. Memories with negative momentum (consistently misleading) get penalized.

Arguments

• context: Current context/query text
• memory_system: Memory storage
• graph_memory: Optional graph memory for entity lookup
• config: Relevance configuration
• momentum_lookup: Map of memory_id -> momentum EMA (-1.0 to 1.0)

pub fn surface_relevant_with_ab_test( &self, context: &str, user_id: &str, memory_system: &MemorySystem, graph_memory: Option<&GraphMemory>, config: &RelevanceConfig, ab_manager: &ABTestManager, ) -> Result<(RelevanceResponse, Option<ABTestVariant>)>

Surface relevant memories with A/B test integration

When an A/B test manager is provided, this method:

1. Assigns the user to a variant (control or treatment)
2. Uses weights from the assigned variant
3. Records impressions for the test

Returns the variant used along with the response for tracking.

pub fn record_ab_click( &self, user_id: &str, memory_id: Uuid, ab_manager: &ABTestManager, ) -> Result<()>

Record a click for A/B testing

Call this when a user interacts with a surfaced memory.

pub fn record_ab_feedback( &self, user_id: &str, positive: bool, ab_manager: &ABTestManager, ) -> Result<()>

Record feedback for A/B testing

Call this when a user provides explicit feedback on a surfaced memory.

pub fn refresh_entity_index(&self, graph_memory: &GraphMemory) -> Result<()>

Refresh the entity index from GraphMemory

This builds an O(1) lookup table from entity names to memory IDs. Call this periodically or when the graph changes significantly.

Performance: O(E * M) where E = entities, M = avg memories per entity Typically completes in <10ms for 1000 entities.

pub fn get_memories_for_entity( &self, entity_name: &str, graph_memory: Option<&GraphMemory>, ) -> Option<HashSet<Uuid>>

Get memory IDs for an entity name using the cached index

Returns None if entity not in index, empty set if entity known but has no memories. Falls back to direct GraphMemory lookup if index is stale or missing.

pub fn entity_index_needs_refresh(&self, max_age_hours: i64) -> bool

Check if entity index needs refresh (older than max_age_hours)

pub fn entity_index_stats(&self) -> (usize, Option<DateTime<Utc>>)

Get entity index statistics

pub fn clear_entity_index(&self)

Clear the entity index (useful for testing or memory pressure)