RAG Glossary - Document AI Terms and Definitions
Comprehensive glossary of RAG, LLM, and artificial intelligence terms. Clear definitions of key concepts to master document AI.
Core RAG Terms
- RAG (Retrieval-Augmented Generation)
- A technique that enhances LLMs by retrieving relevant information from an external knowledge base before generating a response.
- Embedding
- A dense vector representation of text that captures its semantic meaning. Similar texts have embeddings that are close together in vector space.
- Vector Database
- A specialized database for efficiently storing and searching embedding vectors. Examples: Qdrant, Pinecone, Weaviate, ChromaDB, Milvus.
- Chunking
- The process of splitting documents into smaller segments to enable more precise retrieval and respect model context limits.
- Semantic Search
- Search based on meaning rather than exact keywords. Uses similarity between embeddings to find conceptually related documents.
LLM-related Terms
- LLM (Large Language Model)
- A large-scale language model trained on vast text corpora. Examples: GPT-4, Claude, Llama, Mistral.
- Context Window
- The maximum number of tokens a model can process in a single request.
- Hallucination
- Generation of false or invented information by an LLM. RAG reduces this by grounding responses in verified sources.
- Fine-tuning
- Additional training of a pre-trained model on specific data to adapt its behavior or knowledge.
Advanced RAG Techniques
- Reranking
- A step that re-orders search results to improve relevance using cross-encoder models.
- Hybrid Search
- Combination of semantic search and lexical search (BM25) to benefit from both approaches.
- Agentic RAG
- RAG where an AI agent dynamically decides when and what to retrieve, enabling multi-step tasks and complex reasoning.
- GraphRAG
- An extension of RAG that uses knowledge graphs to capture relationships between entities.
- Self-RAG
- A RAG architecture where the model evaluates and critiques its own retrievals and generations.