A vector database acts like a search engine for multimodal data, bringing different sources and modalities together for matching, ranking, and reasoning. section 15.6
Its practical role emerges in AI-powered search: it can provide an information source that grounds a language model with accurate, up-to-date context instead of relying only on generated answers. section 1.0
A vector database can participate in search systems that combine:
- Text, audio, video, metadata, sensory, and temporal modalities as additional context for retrieval and ranking. section 15.6
- Search indexes and language models in retrieval-augmented generation pipelines. section 1.0
- Matching, ranking, and reasoning over information gathered from multiple data sources. section 15.6
Grounding a generative answer
A retrieval-augmented generation system can use a vector database as its knowledge source, retrieve relevant information, and pass that information to an LLM as context for answering a user. section 1.0
Figure 15.17 The convergence of contextual technologies to deliver more intelligent results
Use the convergence diagram to relate vector databases to multimodal search, ranking, reasoning, and other contextual technologies. section 15.6
