Dashboard creation with AI (LLMs) - Examples

[john_martin]

After we saw an example at Perform, we started toying around with creating dashboards via prompts in LLMs. Utilizing Claude, we begged, borrowed, and stole from our Dynatrace consultants and created several dashboards with such, but that had me thinking, what else have other Dynatrace customers have done? 

I've attached a sample.json that we used to give the AI a framework to work from as well as the servicehealth.json that was output from the following prompt: 

I'm an SRE. Build a Dynatrace dashboard for me to help observe the main KPIs that SREs care about. Take a look at the sample Dashboard in the directory as I like that look and feel of that dashboard. Enable auto select for timeseries charts only in visualization settings. For K8s Node Status, make sure the DQL for sourceField uses entity_name and not id. If you build any honeycomb tiles, place them to the left for that row. For any network tiles, use dt.host.net.nic.bytes_rx and dt.host.net.nic.bytes_tx metrics. Make sure all functions used in DQL are correct based on dynatrace official documentation (https://docs.dynatrace.com/docs/discover-dynatrace/platform/grail/dynatrace-query-language) and parsing is correct in the DQL based on dynatrace official DPL documentation (https://docs.dynatrace.com/docs/discover-dynatrace/platform/grail/dynatrace-pattern-language)

[zietho]

I don't want to spoil too much but let's say we are anyways looking into this... 

Important pieces that we ourselves are still trying to figure out at this point is what instructions and validations are needed to create a:

• technically correct
• visually appealing 
• and cost and performance efficient document

So indeed, context is key. And at least think of leveraging the following available pieces: 

• DT MCP / DT app MCP for querying, the semantic dictionary, Strato colors etc., 
• Good examples such as ready-made - or as in your case - your own favorite dashboards. Last time I tried my luck I used the most frequently used 3-5 Ready-mades as input (Kubernetes cluster, Infrastructure, Database, ..).
• DTCTL to as a authentication tool for your LLM/harness and to download/upload the created document.

These would cover a few aspects, namely, validity/validation, visual appeal, and ease of interacting with your tenant. 

At the end all guaranteed that the result can be imported and the queries work, while not having to manually do anything with the tenant. All CLI first, let claude or any other LLM model of your choice handle the grunt work. 

But curious whether and what others have done and what their experiences have been.. !?