Some parts of the JuliaData ecosystem are uber cool, like DataFrames, TidierData, DuckDB, etc. However, they lack robust support for parquet, iceberg, accessing data in ADLS, etc. There are workarounds like using DuckDB for accessing parquet files, but that's not always ideal.
For visualization, there are tons of great libraries like Makie (complex and powerful), VegaLite (very easy to use), and PlotlyLight.
One aspect which is seriously lacking is the ability to create nice web applications. There is GenieFramework (somehow I have always encountered issues with it), then there is Pluto (also a great idea but not a great experience). For static reports, QuartoNotebooks are awesome.
Once you start going deeper into statistical analysis, my experience is hit-or-miss depending upon what I am trying to do. The TimeSeries analysis ecosystem, for example, is fragmented and not as mature.
But with the advent of LLMs, I can easily and quickly write code and create custom functions for just the task I am working on, which I believe would be great for Julia. You can quickly create a custom, performant, pure Julia implementation for the task at hand.
For interacting with LLMs, PromptingTools.jl is awesome.