Glossary

Answer / Output: The final graph stage that represents a response delivered by a future runtime.

Backend execution: Server-side processing of a pipeline using actual documents, models, retrieval infrastructure, and authorization. Not implemented in the reviewed frontend.

BM25: A keyword-based information retrieval ranking technique used conceptually by Ragify's Hybrid Retriever.

Chunk: A smaller segment of a larger document used as a retrieval unit.

Dense retrieval: Search using numerical embedding similarity rather than only keyword matching.

Embedding: A vector representation of content used for similarity operations.

Frontend preview: A user-interface capability, including design and sample-data surfaces, that does not imply connected backend behavior.

Graph: A collection of nodes and directional connections describing a workflow.

Graph validation: Checks of supported structure and configuration performed by the current frontend.

Hybrid retrieval: Combining semantic and keyword retrieval signals into a single candidate ranking.

Ingestion: The process of receiving, parsing, chunking, and indexing source material. Planned in the backend.

LLM: Large Language Model, used conceptually for answer generation.

Node: A configurable element representing a pipeline stage.

Pipeline: An ordered collection of processing, retrieval, and output stages.

RAG: Retrieval-Augmented Generation, where retrieved information contributes context to language-model generation.

Reranking: Re-scoring and reordering retrieved candidates after an initial search.

RRF: Reciprocal Rank Fusion, a method of combining ranked result lists from multiple retrieval strategies.

Session-scoped: State associated with the active editor/browser session rather than durable backend persistence.

Vector store: Data infrastructure intended to store embedding vectors for retrieval; not connected in the reviewed frontend.