A vector database stores and indexes numerical representations of information, commonly called embeddings, enabling efficient similarity searches across large collections of structured and unstructured content. Option D is correct because vector storage and retrieval are foundational capabilities in many Retrieval-Augmented Generation architectures.
Embeddings are multidimensional numerical vectors generated by embedding models. They encode semantic characteristics of documents, queries, or other information, enabling systems to identify related content even when exact keywords differ.
For example, an enterprise knowledge assistant may convert Cisco configuration guides, troubleshooting procedures, and operational manuals into vector embeddings.
When a user asks a technical question, the system generates a query embedding and searches the vector database for semantically relevant information. Retrieved passages are then supplied to the generative model as supporting context.
Option A is incorrect because computational resource allocation is managed by the inference infrastructure or orchestration platform.
Option B is incorrect because the model's context window is determined by its architecture and serving configuration. The retrieval process can select content for that window but does not control its maximum capacity.
Option C is incorrect because updating model weights belongs to training or fine-tuning, not retrieval.
RAG enables applications to incorporate external knowledge without requiring the underlying generative model to be retrained for every information update.
Specific Cisco AITECH Course Guide References:
Customizing AI Models
Generative AI Ecosystem
Exam Objective 1.5: Retrieval-Augmented Generation, embeddings, and vector databases
Cisco explicitly identifies embeddings and vector databases as elements of the RAG examination objective.
===================
Submit