would need to test how well it handles messy real-world datasets and whether the AI picks appropriate chart types consistently. | discoverkit | discoverkit