A user asks about "my last invoice." What should the app add to the prompt?
Why C is the answer
Context based on key fields and intent
Pulling out the key fields and intent lets the app fetch the right data and add it to the prompt, so the model answers exactly.
In detail
- A prompt can be augmented from a user's key fields, terms and intents.
- Here, that means adding this user's invoice details.
- Add only this user's data, which protects privacy and saves tokens.
Use case
An illustrative scenario, not a real incident.
A user asks about "my last invoice". The app finds the user's ID and intent, fetches that invoice, and adds it to the prompt, so the model can answer exactly.
See it in different ways
sequenceDiagram
participant U as User
participant A as Agent
participant I as Vector index
participant T as Tool
U->>A: Question
alt General policy
A->>I: Search documents
I-->>A: Passages
else Live or per-record fact
A->>T: Call with a key
T-->>A: Current value
end
A-->>U: Answerflowchart LR S0["Read the question"] --> S1["Find key fields and intent"] --> S2["Fetch matching data"] --> S3["Add to the prompt"]
Key point: Add context from the user's key fields and intent.
Why the other options are wrong
- A. A higher temperature
- Temperature adds no facts.
- B. Every invoice in the company
- That wastes context, and exposes other customers' data.
- D. A longer list of unrelated examples
- Unrelated examples add noise.