Retrieving Tables and Notebooks for a Data-Question Agent

Reported in interviews at Databricks

Problem statement

How would you find tables and notebooks for an agent answering employee data questions? The catalog is large and full of stale or duplicate assets. Retrieval must be fast, return only a few useful assets, and respect the asking user’s permissions.

Explain indexing, retrieval, asset quality and evaluation.

Show hint

What words would a person asking the question use, and where in each asset could those words possibly appear?

Relevant roles: ML Engineer, AI Engineer, Applied Scientist and LLM Engineer.

Related free problems