Struct ExtractionPipeline 

pub struct ExtractionPipeline<P: ExtractionProvider> { /* private fields */ }

The main extraction engine. Takes raw conversations, extracts structured memories via an LLM, then filters and validates them before storage.

Implementations

impl<P: ExtractionProvider> ExtractionPipeline<P>

pub fn new(provider: P, config: ExtractionConfig) -> Self

pub async fn extract_from_conversation( &self, conversation: &str, ) -> Result<Vec<ExtractedMemory>, ExtractionError>

Call the LLM to extract memories from a conversation, then parse the response and filter by quality threshold.

pub async fn extract_full( &self, conversation: &str, ) -> Result<ExtractionResult, ExtractionError>

Extract memories AND entities from a conversation. Returns the full ExtractionResult including structured entities.

pub fn filter_quality( &self, memories: &[ExtractedMemory], ) -> Vec<ExtractedMemory>

Remove memories below the configured confidence threshold.

pub fn check_contradictions( &self, new_memory: &ExtractedMemory, existing: &[MemoryNode], embedding_provider: &dyn EmbeddingProvider, ) -> Vec<CognitiveFinding>

Check a new extracted memory against existing memories for contradictions using the WriteInferenceEngine.

pub fn check_duplicates( &self, new_memory: &ExtractedMemory, existing: &[MemoryNode], embedding_provider: &dyn EmbeddingProvider, ) -> bool

Check if a new memory is too similar to any existing memory (above deduplication_threshold).

pub async fn process( &self, conversation: &str, existing_memories: &[MemoryNode], embedding_provider: &dyn EmbeddingProvider, ) -> Result<ProcessedExtractionResult, ExtractionError>

Run the full extraction pipeline: extract -> filter quality -> check duplicates -> check contradictions.