I have an experiment at SignalRank. agent.signalrank.com - a full semantic front end to our data. It can use superset dashboards as a source, datawrapper, and snowflake.
yes, the fact Superset is open source and free and pretty much just uses native SQL makes it a very interesting proposition where the client does not want Power BI for whatever reason.
I have just installed Superset on a VM recently to have a poke around.
I think an even bigger question is who we are building BI for and we skip data presentation right to decision-making. Even a perfect BI is just an illustration, it's a person who acts on it is adding value. We used to delegate BI work to analysts, so they know what is going on with the domain and then support some thesis or decision. A good chart used to be the proof-of-work we were digging hard enough. Nowadays the chart is becoming an afterthought or a decision auditing tool if find clever ways to delegate the decisions to "agents".
"there’s just something really weird and unpleasant about writing 20 lines of config to make a scatter plot" 💯 totally agree even ggplot feels a lot more natural
I ported a huge Metabase set of dashboards to Superset this weekend. Including a design handbook Claude did for SignalRank. About 24 hours supervising Codex for the entire thing. Example of 1 dashboard here
Claude can right "nightmare insecure code". It can also write good code. It's all about the context it's operating in--guardrails, knowledge, test harness, etc. In the right context it's more than capable of writing production-grade code. If this isn't something you've personally experienced yet I highly recommend finding out a way to get that experience.
Hi Tristan, yep, if AI is going to be useful in the data warehousing area it is going to have to be able to understand the data warehouse to build analytics. I dont see that happening for a while but I very interested to hear anyones progress in this area.
Apache Superset is an interesting part of this puzzle. The YAML is the only missing piece.
I have an experiment at SignalRank. agent.signalrank.com - a full semantic front end to our data. It can use superset dashboards as a source, datawrapper, and snowflake.
Hi Keith,
yes, the fact Superset is open source and free and pretty much just uses native SQL makes it a very interesting proposition where the client does not want Power BI for whatever reason.
I have just installed Superset on a VM recently to have a poke around.
Codex-> Data -> YAML -> Superset
I think an even bigger question is who we are building BI for and we skip data presentation right to decision-making. Even a perfect BI is just an illustration, it's a person who acts on it is adding value. We used to delegate BI work to analysts, so they know what is going on with the domain and then support some thesis or decision. A good chart used to be the proof-of-work we were digging hard enough. Nowadays the chart is becoming an afterthought or a decision auditing tool if find clever ways to delegate the decisions to "agents".
"there’s just something really weird and unpleasant about writing 20 lines of config to make a scatter plot" 💯 totally agree even ggplot feels a lot more natural
I ported a huge Metabase set of dashboards to Superset this weekend. Including a design handbook Claude did for SignalRank. About 24 hours supervising Codex for the entire thing. Example of 1 dashboard here
https://data.signalrank.com/public/dashboard/the-signalrank-index
How do you maintain the nightmare insecure code that Claude creates?
Claude can right "nightmare insecure code". It can also write good code. It's all about the context it's operating in--guardrails, knowledge, test harness, etc. In the right context it's more than capable of writing production-grade code. If this isn't something you've personally experienced yet I highly recommend finding out a way to get that experience.
Hi Tristan, yep, if AI is going to be useful in the data warehousing area it is going to have to be able to understand the data warehouse to build analytics. I dont see that happening for a while but I very interested to hear anyones progress in this area.