Kurt Buhler
@kurtbuhler • 6,422 subscribers
Data Goblin.
Videos
An example of a dashboard in a live artifacts, created in a few simple prompts. It's connected to a semantic model in Power BI, querying in DAX. Slow (fixable, it seems) but interesting results. Also seems that sharing isn't available yet for live artifacts that i can tell?
Kurt Buhler94,777 görüntüleme • 2 ay önce
Here's a quick example of a Data App in Fabric, querying a semantic model. Creating a prototype like this takes maybe 10-20 minutes using the tools from Microsoft and a coding agent. Creating something you can share with users takes a lot longer of course. :)
Kurt Buhler24,494 görüntüleme • 1 ay önce
If you make Power BI reports, I think you might enjoy stepping outside of Power BI to experiment with coding agents and dataviz libraries like Vega, Altair, or d3.js. Here's an example of a dashboard I made with FitBit data. While making the pbir-cli (testing starts this week, thanks to everyone who showed interest!), we learned a lot. Chiefly: we've found it's really, really hard to get an agent to work well with Power BI reports!! Power BI reports are built around abstraction and different intersecting layers. Themes, metadata, extensions, the model... all of it built for a low-code SSBI experience (clicky-clicky-draggy-droppy). That's fine for you and me (mostly). For a coding agent, though, it's inflexible... maybe even hostile. Even with the right tools and context, you simply can't get a Power BI report to reliably do _exactly_ what you want; it's just not meant for that. If you experiment a bit with Vega, Altair, or d3.js... you'll see what I mean; the contrast is pretty clear. This isn't necessarily a bad thing; unlimited flexibility may lead to inconsistency, and there's creativity in constraints. But still, I've been wanting to explore visualization without these constraints... My experiments beyond Power BI reports have been interesting and given me a lot to think about. The most promising approach so far: viz-as-code with custom embedded visuals, pulling from local or remote data sources. This stepcount dashboard is one such example... made in just a few hours between other tasks. Materializing a dashboard like this from natural language wasn't really viable a year ago. Project out another 12–18 months and it's certainly difficult to imagine what the future looks like... That said, reports and dashboards in an enterprise context aren't just visuals. Governance, distribution, integration, etc... years of individual and organizational investment. All of that makes change have a speed limit; we won't be plugging in custom embedded dashboards, overnight. But if you're curious, I do think it's a good investment to step outside Power BI and experiment. Investments you may even be turned into custom visual designs, anyway :). As Havens Consulting put it recently to me: it's time to multiclass...
Kurt Buhler28,475 görüntüleme • 3 ay önce
Daha fazla içerik yok.