How do you work with Databot? ("good practices")

Positron
1

Hey everyone,

I'm quite enchanted with Positron and very impressed by the capabilities of Databot. I really like the insights it helps uncover quickly, the export features (to notebooks and reports) and the right level of interactivity and oversight.

Generating the insights is very fast so the the QA and curation/priorisation of insights and fine tuning of graphics etc for the reports becomes relatively more part of the work (for the driver in front of the screen).

I'd love to hear/exchange some practical experiences in using this bot.

For example, I find myself copy and pasting the most interesting insights and plots into a separate .md file (using visual mode), which I then reference or copy before the final report is created. so how do you use this tool, what kind of usage patterns and tricks have you uncovered, that fit your workflow?

  • Also, if you like to share: how do your Databot.md files look like? :slight_smile:
    (Is there a global Databot.md file, that could be shared across projects?)
  • and how do you go about in checking the results? Do you try to read the whole code, do you plausibility-check (depending on the stakes), do you hand code some alternatives?

For long Databot Chats some kind of generated or editable outline for the various steps of the analysis would/could also be really nice for navigation.

2

Not exactly about DataBot, but Andrew Heiss' talk addresses the Positron environment.

I have found that i need to go through DataBot output chunk by chunk. There are subtle changes in logic that I need to propagate through an entire file, so I ask DataBot to write a qmd file with all the code. That seems to consume a big number of tokens. I haven't automated fixes at all.

All of that said, DataBot's roungh-and-ready, cowboy-style products end up pointing me in unexpected and very useful directions.

1 Like