Thinking in Style: a Report Ninja update

Modified on Tue, 23 Sep at 4:55 PM

"An update from Visokio’s satanic mills of code,

where deep in the mountain

developers hammer and forge new AI-driven powers

for Omniscope’s Report Ninja."

Report Ninja was first released as a labs feature about 6 months ago and introduced an AI companion for chatting to Omniscope to build dashboards. Since then we've added a few things:

Instant Dashboard, a one-shot auto-generated dashboard for a given dataset;
Streaming, so incoming messages flow in front of your eyes - no long waits with nothing happening;
Thinking, so you can see the inner workings of the AI's "mind", and get a better feeling of what's happening while you wait for the outcome;
Local LLMs supported, self-hosted for privacy and control;
Data Q&A: ability to answer simple data questions visually.

We're now wrapping the latest iteration which brings the ability for the AI to use style and layout in its dashboard designs. So now your reports not only compute smartly but look smart too. Plus we've expanded a little on the "Data Q&A" capabilities. Here the views are laid out better, and UI styles, colours and fonts have been customised automatically.

The latest gpt-5 models perform extended "thinking" (similar to previous o-series models), and

you can control the thinking effort and verbosity. Low thinking tends to give a good outcome, minimal makes it respond instantly although perhaps less intelligently. For a dashboard assistant, we find OpenAI models to think for too long on medium or high. Here's our recommended settings:

Here's a nice example of the 3 verbosity levels. We like "low"!

And finally, simple data Q&A means you can ask a question and get a visual answer. "How have matches trended over time?" results in a time-series chart. What's new here is that the AI is now able to apply filters and understand your data better, resulting in better answers.

Some of these features are still pre-release. If you want access to the closed beta, drop us a line - support@visokio.com - and let us know what you think.

Older articles:

Data used here was thanks to Kaggle.