You’re asked to describe how you would build an LLM-based agent that can converse with a user (e.g., an interviewer) and answer questions using an internal knowledge base.
Requirements
-
The agent should use
retrieval-augmented generation (RAG)
over a document corpus.
-
It may call
tools/APIs
(search, calendar, database lookup) when needed.
-
Discuss safety/guardrails and evaluation.
Follow-up
-
Where does the
randomness
in LLM outputs come from?
-
What does the
temperature
parameter do, and how do
top-k
/
top-p
relate?