GLOSSARY
Vector database
A vector database stores and searches numerical representations of content to find similar records. These representations are often produced through embeddings. This allows retrieval of relevant passages even when a question uses different words from those in the stored document.
When a similar proposal supplies the wrong answer
A sales team searches for an established response to a customer requirement. An older proposal matches closely but contains outdated prices and exceptions agreed with a particular customer. Similarity explains why it was retrieved. Its present suitability also depends on validity, provenance and access permissions.
The collection therefore needs maintained information such as product version, applicable period and permitted audience. Changes and deletions must reach the search index as well. Otherwise, withdrawn statements can reappear as seemingly reliable knowledge. Poor information maintenance may then reproduce the same mistake across many later proposals.
Assessing retrieval quality
Useful test questions include expected sources, outdated alternatives and prohibited results. Measure relevant passages found and significant omissions, then assess the generated answer separately. A vector database can form part of retrieval-augmented generation. Whether it justifies its cost depends on the information collection and actual retrieval task; simpler search methods belong in the comparison too.