retrieval-augmented-generation-architect-module

memory entryAtomic unit of stored information in the memory system, containing content plus metadata

retrieval augmented generationParadigm that augments language model generation with information retrieved from external knowledge sources to

dense passage retrieverBi-encoder model trained on query-passage pairs to produce jointly learned dense embeddings for retrieval

dense retrievalNeural retrieval using learned embedding models to encode queries and documents into dense vectors for similar

memory storePersistent or semi-persistent storage layer holding memory entries across retrieval operations and sessions

mmr001Modality type: classification of data representation forms a retrieval system handles — text, image, table, co

mmr002Multi-modal embedding model: neural network architecture that projects content from different modalities into

mmr003Cross-modal retrieval: searching for content in one modality using a query from a different modality, such as

mmr005Non-text chunk processor: component that parses, extracts, and encodes non-text content from documents — handl

mmr009Multi-modal index: index structure supporting heterogeneous chunk types — text vectors, image features, table

mmr027Multi-modal context builder: constructing the generation-augmenting context by assembling retrieved chunks of

rrf fusion scoreThe sum of reciprocal rank values for a given document across all contributing rank lists; the document's posi

vector databaseSpecialized storage system indexing embeddings with efficient similarity search capabilities (Pinecone, Weavia

embedding modelNeural model converting text into dense vector representations optimized for retrieval similarity tasks

hybrid searchRetrieval combining multiple search paradigms (dense/sparse, vector/keyword) with score normalization or late

mmr011Heterogeneous retrieval engine: unified retrieval layer that queries across mixed content types, resolving mod

multi modal ragRAG system that retrieves and augments generation with images, tables, code, or other non-text modalities alon

colbert approximate searchApproximate nearest neighbor indexing on per-token document embeddings enabling efficient retrieval of candida

corrective ragRAG architecture with explicit error detection: low-quality retrieval triggers fallback to web search or re-re

end to end rag evalSystem-level evaluation measuring whether RAG outputs are factually correct, complete, and grounded in retriev