4 Comments
Dec 1, 2022·edited Dec 1, 2022Liked by David Jayatillake

Awesome article David! It’s great to see how AskEdith performs on real-life data.

Like you, I like “Approach 2” (both in the context of natural language and in general):

1. It’s actually harder for data/analytics engineers to do Approach 1 than Approach 2. Building “one big table”s with all the right joins and granularity is tricky. There are so many ways of slicing and dicing data, so it’s nice to let (for example) LookML generate joins for you.

2. For at least the next few years, Natural Language won’t be the main way we interact with data. It’ll exist alongside “traditional” BI tools. And those tools all have their own definitions of metrics and dimensions (basically, a semantic layer even if it’s not an open one like Transform or dbt’s). You really want them to match.

(Disclaimer: not exactly an unbiased perspective, lol. I’m building Delphi[0] which helps data teams who struggle with self-service by adding Natural Language capabilities to the modern data stack. And it very much takes Approach 2.)

[0] https://delphihq.com

Expand full comment