Skip to main content

AI Innovation

GENAI DASHBOARDS

GenAI Dashboards

Type a sentence. Get a dashboard built from your own data.

One written request becomes a board of KPI tiles, charts and tables. The model decides what the board looks like. The code computes every number on it.

Exploited findings by system, and how they split by severity

Exploited, outstanding

312

Systems affected

3

By system

GRACE224
AREA66
ALLOY22

By severity

Critical119
High137
Medium56

Top findings

CVESystemsAssets
CVE-2025-31324341
CVE-2025-24813227
CVE-2024-21762219

Illustrative board. Synthetic values.

The rule

The model picks the shape. Code computes every number.

There is no path in which model output becomes a figure, a label or a row. That is what makes a generated dashboard safe to put in front of an Authorizing Official.

01

Your sentence

Name the data you want and how to cut it. Say by system for a per system breakdown, or name a shape: line, pie, scatter, heat map.

A request

02

The plan

One model call returns a plan of tiles: dataset, chart type, group by, split, filters, period, title. Each tile is checked against the vocabulary the model was handed. A tile that does not parse is refused, never repaired.

Validated tile plan

The only model step

03

The counting

Every tile is filled by one of a handful of counting primitives, scoped to the systems you are cleared for. Which primitive fills a tile is decided from the column, never by the model.

Real figures

04

The board

Drawn, and yours to keep: save it, share it, or pin it as the opening board for a role so the first thing an executive sees is the board you chose.

Saved, shared, pinned

Thirteen shapes

Ask for a shape by name

If the data cannot feed the shape you asked for, the board says why and draws the shapes the data can feed. A line over a text column, a scatter of a status: it tells you rather than pretending.

KPI

One headline number

Bar

Counts by category

Stacked bar

Counts by category, split again

Donut

Share of the whole

Pie

Share of the whole, as slices

Treemap

Share of the whole, as sized boxes

Heatmap

Density across categories

Radar

Categories around a wheel

Table

The rows themselves

Line

A count over time or order

Area

A line with the volume filled

Histogram

How a number is spread

Scatter

Two numbers against each other

Three ways onto the page

The opening board

Loads before you type anything: one headline number per dataset and the most telling tile of each. Built entirely in code, or pinned by a curator for your role.

A dataset summary

Press a dataset's button or ask for a summary. A backbone of tiles is built in code from the dataset's roles; the model may add a few more on columns it judges telling.

A plain request

Anything else. The planner reads the sentence and returns the tile plan above. Example prompts on the page show the phrasings that work best.

From Ask Alloy

An answer in the drawer can become a board with the same filters, without the model being asked again.

It draws from every dataset published in the Dataset Builder that your role may open. Adding a data source is a Builder action, not a code change.

Seen in the app

The real thing, on synthetic data

One request, one board. Open versus total, status mix, severity by status and a per system heat map, computed in code over 1.5 million synthetic findings across 120 systems. Every number on the board is counted, never generated.

What you can do

  • Name the data you want, for example vulnerabilities, compliance, assets or ATO, so the right collection is selected

  • Say by system or per system for a per system breakdown

  • Save a board you like, share it, or pin it as the opening view for a role

  • Confirm any figure against the underlying page before you brief it

Straight answers

Can the model put a wrong number on the board?

No. It never produces one. It chooses the dataset, the columns and the shape, and the platform counts. If its choice does not fit the data, the tile is refused rather than repaired.

What can it draw?

KPI tiles, bar, stacked bar, donut, pie, treemap, heatmap, radar and table over any groupable column; line and area over an ordered one; histograms over numbers; scatter over two numbers. Quick answers to a focused question, not a general purpose report builder.

Does my data leave the environment?

No. The planner is a local model inside your own boundary, and the counting is done by the platform against your database.

Is it still being developed?

Yes. Handling a wider range of free form questions depends on model capacity and available GPU in the environment, and that is an active area of work.

Ask for a board. Get one.

Ask for a demo and type your own question at a synthetic dataset. Then pin the result as your opening board.