Are the complex requests in regards to getting/manipulating the data, or in the final visualization. I'm trying to build a desktop tool similar to Tableau but with the ability to easily drop down into python and do data manipulation with the pydata stack. Intended audience is the new breed of quantitative analysts/data scientists.
Both. While tableau has a visual query builder, sometimes the queries are too complex. For example they may need self-joins, non-standard join criteria, temporary tables, etc. So ultimately, the tableau guy has to ask a programmer or database person to create a script to use as the tableau source.
One example that involves both is when the visualisation needs a contiguous date range in the data, but the data is missing some dates. As a programmer it is easy to just loop through a date range and put 0 where there is no data.
There are also lots of limitations in terms of the visualisations. The end result was that multiple visualisations had to be created to show something that a programmer would be able to create as one visualisation.
I would say mainly visuals, think Photoshop meets Excel meets Javascript. One of the founders of Tableau founded Pixar studios so you can imagine that graphics are important. But it also has some great data options.
