I built BigAsk, a self-deployed web interface for exploring BigQuery data by asking questions in natural language. It’s a fairly thin wrapper over the Gemini CLI meant to address some shortcomings it has in overcoming data querying challenges organizations face.

I have a few friends who work in roles where much of their time is spent fulfilling requests to fetch data from internal databases. I’ve heard it described as a “necessary evil” of their job which isn’t very fulfilling to perform. Recently, Google has released some quite capable tools with the potential to enable those without technical experience using BigQuery to explore the data themselves, both for questions intended to return exact query results, and higher-level questions about more nebulous insights that can be gleaned from data. While these certainly wouldn’t completely eliminate the need for human experts to write some queries or validate results of important ones, it seems to me like they could significantly empower many to save time and get faster answers.

Unfortunately, there are some pretty big limitations to the current offerings from Google that prevent them from actually enabling this empowerment, and this project seeks to fix them.

One is that the best tools are available in a limited set of interfaces. Those scattered throughout the already-lacking-in-user-friendliness BigQuery UI require some foundational BigQuery and data analysis skills to use, making their barrier to entry too high for many who could benefit from them. The most advanced features are only available in the Gemini CLI, but as a CLI, using it requires using a command-line, again putting it out-of-reach for many.

The second is a lack of safe access control. There's a reason BigQuery access is typically limited to a small group. Directly authorizing access to this data via the BigQuery UI or Gemini CLI to individual users who aren't well-versed in its stewardship carries large risks of data deletion or leaks. As someone with experience working professionally with managing cloud IAM within an organization, I know that attempts to distribute permissions to individual users while maintaining a limited scope on them also requires considerable maintenance overhead and comes with it’s own set of security risks.

BigAsk enables anyone within an organization to easily and securely use the most powerful agentic data analysis tools available from Google to self-serve answers to their burning questions. It addresses the problems outlined above with a user-friendly web interface, centralized access management with a recommended permissions set, and simple, lightweight code and deployment instructions that can easily be extended or customized to deploy into the constraints of an existing Google Cloud project architecture.

Code here: https://github.com/stevenwinnick/big-ask

I’d love any feedback on the project, especially from anyone who works or has worked somewhere where this could be useful. This is also my first time sharing a project to a forum, and I’d value feedback on any ways I could better share my work as well.