Questions about exact part numbers fail in vector search. What improves this?
Why D is the answer
Hybrid keyword and vector search
Keyword search handles exact identifiers.
In detail
- Keyword search handles exact identifiers.
- Hybrid search uses both.
Use case
An illustrative scenario, not a real incident.
A search for part "A-77B" fails on vectors, and keyword search finds it.
See it in different ways
flowchart LR A["Ingest and chunk"] --> B["Embed"] B --> C["Index"] D["Agent query"] --> C C --> E["Rerank"] E --> F["Context for the agent"]
stateDiagram-v2 [*] --> Parse: layout-aware Parse --> Index Index --> HybridSearch: keyword and vector HybridSearch --> Permissions: filter by user Permissions --> Rerank Rerank --> [*]
flowchart TD
Q{"What is searched for?"}
Q -->|"Exact identifiers"| A0["Keyword search"]
Q -->|"Meaning"| A1["Vector search"]Key point: Keyword search finds exact identifiers.
Why the other options are wrong
- A. A larger temperature during retrieval
- Temperature has no effect on retrieval.
- B. Longer system prompts with the part list
- A list in a prompt does not scale.
- C. A smaller number of documents in the index
- Fewer documents lose knowledge.