# 2026 Hackathon Winner: Visage, a Looker Plugin for Prompt-Driven Custom Data Visualisations

URL: https://rittmananalytics.com/blog/2026/11/18/visage-custom-looker-charts-from-a-prompt
Description: Visage turns a plain-English description of a chart into a custom Looker visualisation, then renders it with real Looker data, screenshots it and runs it through 22 checks before it is published.
Author: Alex Caldwell
Published: 2026-11-18T16:56:21+00:00
Modified: 2026-11-18T16:56:21+00:00

---

![2026 Hackathon Winner: Visage, a Looker Plugin for Prompt-Driven Custom Data Visualisations](https://rittmananalytics.com/api/public/og-image?src=https%3A%2F%2Fi0.wp.com%2Fdtdtemp8.wpcomstaging.com%2Fwp-content%2Fuploads%2F2026%2F11%2Fvisage-fig1-prompt-to-chart.png%3Ffit%3D1800%252C697%26ssl%3D1)

Visage is a tool I built that turns a plain-English description of a chart into a custom Looker visualisation. You describe what you want to see, Claude Code writes the JavaScript file, and Looker loads it like any other custom chart. You can browse the code on [GitHub](https://github.com/alexander-caldwell/visage) or try every chart in the [live gallery](https://alexander-caldwell.github.io/visage/).

![An example prompt passing through build and check steps to become a Looker waffle chart](https://rittmananalytics.com/api/public/og-image?src=https%3A%2F%2Fi0.wp.com%2Fdtdtemp8.wpcomstaging.com%2Fwp-content%2Fuploads%2F2026%2F11%2Fvisage-fig1-prompt-to-chart.png%3Fresize%3D1024%252C397%26ssl%3D1)

Figure 1: From a plain-English prompt to a checked Looker waffle chart (example prompt, sample data)

Writing the file is only half the job. A custom chart can look fine in code and still show up blank in Looker, so before I publish one I render it with real Looker data, screenshot it and run it through a set of checks.

Below, I explain why custom charts break, how Visage checks them and how to add one to your own Looker instance.

## Where it started

A large retail client wanted a treemap of their RFM (recency, frequency, monetary) customer segments. They wanted to see two things at a glance: how many customers sat in each segment, and how much those customers spent. Box size would show the first and colour the second.

A treemap isn’t one of Looker’s built-in charts. The usual route is to write one yourself against Looker’s [custom visualisation API](https://docs.cloud.google.com/looker/docs/custom-vis-api-reference), which lets developers build and register their own charts in JavaScript. At the time, we gave the client a different chart that someone else had already built and shared online. It worked, but it wasn’t what they had asked for.

With Visage, the treemap they wanted is now a two-sentence request: *“A treemap of customer segments. Box size shows the number of customers and colour shows total spend.”* Visage builds the chart to that description, then renders and checks it before it is published. Any chart that isn’t in the gallery yet starts the same way. The result uses a light-to-dark colour scale so spend is easy to compare across segments, and a caption explains what size and colour mean.

![Visage treemap with box area showing revenue and colour showing margin](https://rittmananalytics.com/api/public/og-image?src=https%3A%2F%2Fi0.wp.com%2Fdtdtemp8.wpcomstaging.com%2Fwp-content%2Fuploads%2F2026%2F11%2Fvisage-fig2-treemap.png%3Fresize%3D1024%252C676%26ssl%3D1)

Figure 2: The Visage treemap on sample data. Box area shows one measure (revenue) and colour shows a second (margin). The grey box has no margin value.

I built the first version of Visage during a six-hour internal hackathon: a test harness, automated checks and five charts. It won, which I mention only in passing. I’ve kept developing it since, adding chart families, improving the charts and changing how they’re hosted.

## Why custom charts break

A custom Looker chart is a single JavaScript file. You host it, register it in Looker Admin or your LookML project’s manifest, and Looker loads it directly as a script, with no build step or bundler.

Most failures are silent: the tile is blank, and the logs don’t say why or even show that anything went wrong. Each attempt at a fix means registering the file again, deploying it and reopening the Explore, so a small bug can eat an afternoon.

I ran into five causes of a blank tile:

1.  **Re-mounted tile.** Looker sometimes rebuilds a tile and gives the chart a new, empty container. If the chart keeps drawing into the old one, nothing appears.
2.  **Zero-size tile.** Looker can ask a chart to render before the page layout is ready. A chart that sees zero width or height and doesn’t try again stays blank.
3.  **Trusted Types.** Some Looker instances enforce this browser security feature, which blocks operations such as `innerHTML`. Code that looks normal fails without warning.
4.  **Missing headers.** GitHub Pages doesn’t send the cross-origin headers Chrome needs for this way of loading a file, so Chrome refuses it without a clear error.
5.  **No `done()` call.** Looker waits for the chart to call `done()` when it has finished rendering. If it never does, the tile never finishes loading.

None of these show up when you read the code. You find them by rendering the chart and looking at it, which is what Visage does.

## How Visage checks a chart

Every chart goes through the same seven steps before it is published:

1.  **Check what exists.** Visage looks at the existing charts and the data they need. Often an existing chart can take a new option instead of a new file.
2.  **Generate the file.** It starts from the chart family’s shared code and controls, so every chart begins from the same tested base.
3.  **Load real data.** The chart is tested with data captured from a real Looker Explore.
4.  **Render and screenshot.** A headless browser loads the chart the way Looker does and takes a screenshot.
5.  **Look at the screenshot.** Claude reviews the image for problems the logs won’t show, such as a blank or wrongly drawn chart.
6.  **Run the checks.** The chart must pass every automated check. If any fail, it goes back to step 4.
7.  **Publish.** A separate workflow publishes the file, its gallery entry, the docs and a tagged release.

![Visage build loop: seven steps, with a loop back to rendering when any check fails](https://rittmananalytics.com/api/public/og-image?src=https%3A%2F%2Fi0.wp.com%2Fdtdtemp8.wpcomstaging.com%2Fwp-content%2Fuploads%2F2026%2F11%2Fvisage-fig3-build-loop.png%3Fresize%3D1024%252C361%26ssl%3D1)

Figure 3: The Visage build loop. Any failed check sends the chart back to the render step.

Every visual bug I found came from step 5, looking at the screenshot: a hover band drawn as a solid black block, two dots hiding each other where their values were equal, and an axis label reading “\-0”, a number nobody has ever needed.

Each automated check comes from something that went wrong in a real Looker instance. **Rendering checks** put the chart through the conditions that broke earlier versions: re-mounted and zero-size tiles, light and dark dashboards, narrow tiles, and queries that return no rows or only nulls. **Source checks** scan the file itself and fail any chart that uses `innerHTML` or picks colours by row position.

A chart with an axis has to pass 22 checks. When I add a new check, every chart is tested against it from then on, so a fixed bug is caught if it returns.

## Rules for readable charts

Some problems aren’t bugs: the chart works, but it’s hard to read. Each of these rules was added after I saw that happen.

1.  **Follow the dashboard, not the laptop.** Charts read their tile’s background, not the viewer’s dark-mode setting, so a dark chart never lands on a light dashboard. Dark mode has its own colour scale so every shade stays visible.
2.  **Shorten labels from the middle.** Long labels wrap first, then shrink, then get shortened, and are dropped only as a last resort. Cutting the middle rather than the end keeps “Phase 2” and “Phase 3” distinct. The tooltip always shows the full text.
3.  **Colour by value, not by row order.** An early treemap assigned colours by row order, so filtering out one segment changed the colour of every other one. Colours now follow the value or category, so they hold when data is filtered.

## The charts

Visage now has a suite of 19 ready-built charts covering the common needs: comparing categories, showing how values are spread, breaking a total into parts, trends over time and single KPIs. You can try them all in the [live gallery](https://alexander-caldwell.github.io/visage/) or read the code on [GitHub](https://github.com/alexander-caldwell/visage).

If the chart you need isn’t there, we can build it to your spec. It goes through the same build loop and checks as every other chart before it reaches your dashboards.

Charts in a family share controls and code, which keeps them consistent. Variations are options rather than new files: stacked, 100% stacked and lollipop columns all come from `grouped_column.js`. The code that stops tiles going blank lives in one place and is copied into every chart at build time. Each chart is still a single file with no dependencies, and a fix made once reaches all of them.

![Dumbbell, column and bullet bar charts from the Visage gallery](https://rittmananalytics.com/api/public/og-image?src=https%3A%2F%2Fi0.wp.com%2Fdtdtemp8.wpcomstaging.com%2Fwp-content%2Fuploads%2F2026%2F11%2Fvisage-fig4-comparison-charts.png%3Fresize%3D1024%252C513%26ssl%3D1)

Figure 4: A few charts from the Visage gallery: dumbbell, column and bullet bar

## Adding a chart to Looker

There are two ways to add a chart, depending on whether you’re trying it out or using it in production. Either way, the chart file holds code only, no data. It runs in the viewer’s browser inside Looker, draws the query results Looker passes to it, and makes no network calls of its own, so your data doesn’t leave Looker. The [hosting guide](https://github.com/alexander-caldwell/visage/blob/main/docs/hosting.md) covers both options step by step.

### Option 1: Register it by URL

In Looker, go to **Admin \> Platform \> Visualizations \> Add Visualization** and enter the chart’s ID, a name and its jsDelivr URL:

```
https://cdn.jsdelivr.net/gh/alexander-caldwell/visage@2/treemap_dual.js
```

The chart then appears in the Explore visualisation picker, with nothing to commit or deploy. `@2` follows the latest 2.x release, so fixes reach you without editing the registration. Pin an exact tag, such as `@v2.0.0`, if the file must never change.

### Option 2: Add it to your LookML project

Copy the file into your LookML project and reference it in the [manifest](https://docs.cloud.google.com/looker/docs/reference/param-manifest-visualization). Looker then serves the chart itself, so there’s no external CDN or cache to manage, and the chart is versioned and deployed with the rest of your LookML. **I recommend this for production**, because you control exactly when the chart changes.

## Which chart do you need?

If there’s a chart you’ve wanted in your Looker dashboards and couldn’t get, tell us. It might be the next one we build.

## Interested? Find out more!

Rittman Analytics is a boutique data analytics consultancy that helps ambitious, digital-native businesses scale-up their approach to data, analytics and generative AI.

We’re authorised delivery partners for Google Cloud along with Oracle, Segment, Cube, Dagster, Preset, dbt Labs and Fivetran and are experts at helping you design an analytics solution that’s right for your organisation’s needs, use-cases and budget and working with you and your data team to successfully implement it.

If you’re looking for some help and assistance with your AI initiative or would just like to talk shop and share ideas and thoughts on what’s going on in your organisation and the wider data analytics world, [contact us now](https://calendly.com/markrittman/initial-discovery-call-with-mark-rittman) to organise a 100%-free, no-obligation call, we’d love to hear from you!
