Hi everyone !
I’m looking for a solution that will allow me to limit responses only to the data that I have provided to the vector database. I use Pinecone.
Scenario:
If I ask a question related to the vector database. Something strictly related to my company. This will answer me and provide sources.
If I ask a general question, e.g. How many Start Trek films have been made?
This will give me an answer, e.g. I’m sorry, but the given document data does not provide any specific information.
This is more of a general prompt-engineering question and not a Pinecone question per se, but I can point you in the right direction.
Those types of constraints or “guardrails” can be included in the prompt directly. There isn’t a precise mechanism to specify this behavior, but something along these lines generally works:
You are a Human Resources representative for a company.
Use the provided company information to answer the user's question.
To answer the question, you will search the information and respond with the answer based on evidence you find within the provided information.
If you don't find the answer, just say that you don't know, don't try to make up an answer.
If the question is too vague, then ask the user to clarify the question.
Company information:
<company info here -- typically the result of the vector search>
Question: Can you explain the PTO policy?
Answer:
Of course, you can tailor these instructions, add or remove, reword, etc. as needed to tune the behavior, persona, additional guardrails, etc you need.
In my prompt system i used several techniques from this site