Hello everyone,
I’m attempting to figure out how TableFirst, a tool I’m working on, might fit into processes that already employ dbt.
The idea is to let non-technical users describe the table they want in natural language, and then have the system work from existing warehouse objects and dbt models. Concretely, we plan to integrate with dbt so that TableFirst can use dbt models and sources as inputs. The tool would look at the graph, metadata and lineage, select candidate models and tables, and then generate the transformations needed to produce the requested table.
A simple example would be an HR or ops stakeholder asking for a weekly payroll summary by department and region. Today this might mean someone on the data team writing a custom query on top of your existing dbt models. With TableFirst, the flow we are aiming for is: the stakeholder describes the table, the system proposes how to combine relevant dbt models and sources, and then creates a derived table that can be refreshed on a schedule.
I am particularly curious how this lines up with how you already work:
-
After you create dbt models, how are new derived tables usually created today?
For example, do analysts mostly write ad hoc queries and dashboards on top of dbt models, or do you tend to create new dbt models for each new “final” table? -
Would a natural language layer that suggests new derived tables on top of existing dbt models be helpful, or would it mostly get in the way of the structure you are trying to maintain?
-
Are there specific risks or anti-patterns you would worry about if a tool is automatically proposing transformations on top of dbt models?
If this sounds interesting, I can share a short demo video and would really appreciate candid feedback on where this kind of tool could be useful and where it would clearly conflict with how you want dbt projects to be managed.