Chart intelligence · Built to license

AI that gets data visualization right.

We build the training data, evaluation models, and critique agents behind AI systems that generate charts. Licensed to model labs, chart libraries, BI vendors, and AI-native data products.

One of AI's jagged edges

AI can generate charts. But not yet the right chart - reliably, defensibly, every single time.

LLMs trained on the open web default to whatever they've seen most often — usually generic line, bar, or pie. They miss finance-canonical waterfalls. They misread data shapes. They scale axes deceptively. They produce charts that look fine to a generalist and embarrass an expert.

The gap exists because there's no training data that captures expert judgment on chart quality. Public datasets cover chart-type classification at best. Nothing captures the dense, opinionated, domain-specific knowledge that distinguishes an exec-grade chart from a default chart.

The pattern isn't random. It follows the contours of what training data does and doesn't cover. Charts have a high ceiling — domain conventions, perceptual rules, audience-fit — but a low and uneven floor in current models. Address the data, and the edge starts to smooth.

We're closing that gap.

Wrong chart type for the data shape
Misleading scales and visual encodings
No domain-canonical chart vocabulary
Generic styling, no exec-grade polish
Annotations missing or wrong
No critique loop, no self-correction
What we sell

One commercial asset, backed by proprietary production infrastructure.

The Chart Quality Dataset is our offering. Everything else — agents, engines, the public game — exists to produce it. We don't license the production stack. We license the output.

The product

The Chart Quality Dataset

The first comprehensive corpus of expert judgment on chart quality, designed for fine-tuning and evaluation of AI systems that generate charts.

  • Pairwise preferences — this chart vs. that chart, with rationale
  • Expert critiques of generated charts — errors, fixes, alternatives
  • Domain coverage — finance, marketing, sales, ops, science
  • Multimodal — chart specs (Vega-Lite, Plotly, code) + rendered images + critique text
  • Versioned, documented, evaluation-ready
Licensed as one-time sale or multi-year subscription. Optionally bundled with a fine-tuned model trained on it. Sold to model labs, chart libraries, BI vendors, and AI-native data products.
Production stack

How the dataset gets built — and why it stays defensible.

Three components produce the dataset in concert. Two stay internal — they're the IP that makes the asset uniquely ours to grow and update. One is public — it builds the community of expert judges.

Internal IP
Chart Creation Engine
Generates structured chart variants across data shapes, domains, and convention sets. Feeds the critique engine and human labelers with a diverse sample stream.
Internal IP
Chart Critique Engine
Evaluates charts against domain-specific quality criteria. Produces structured critiques used as training signal and as a seed for human review.
Public
The Data Viz Game
Public site where domain experts and viz enthusiasts judge real charts. Continuously refreshes the dataset with new human labels at scale.
Who licenses our work

Built for the companies whose products generate charts.

If your product produces charts — directly, through an LLM, or through a customer's workflow — chart quality is your retention metric. We sell the asset that improves it.

Buyer Why they license
AI model labs Improve multimodal chart capability without spending years building specialized training data. Anthropic, OpenAI, Google, Mistral.
Chart library companies Ship AI-powered chart selection, review, and auto-fix as a premium tier. Highcharts, Plotly, AG Charts, ECharts, Vega.
BI & analytics platforms Add chart-quality intelligence to existing copilot and agent features. Tableau Pulse, Sigma, ThoughtSpot, Power BI, Looker.
AI-native data products Differentiate on output quality. Julius, Hex, Wren, Defog, Numbers Station.
Office productivity vendors Embed chart intelligence in Excel, Sheets, Slides, and Acrobat AI features. Microsoft, Google, Adobe.
AI vertical operators Domain-canonical chart generation without building chart logic in-house. AI CFO, AI marketing, AI sales, AI ops products.
Why ChartBoss

A specialised team building one asset, exceptionally well.

Not a general AI consultancy. Not a chart vendor. We do one thing: the data layer behind better AI-generated visualizations. That focus is the moat.

01
Two decades of data visualization depth
Built across analytics product design, BI, and AI-native reporting. We know the conventions, the failure modes, the domain idioms — and we know which generalize.
02
Working on a known jagged edge
Charts are one of AI's documented jagged edges — uneven exactly where it matters. We're systematically reducing the failure rate with JtbD-grounded taxonomy, evaluation rigor, and targeted data. The methodology generalises; charts are the demonstration case.
03
Domain-specific coverage from day one
Finance, marketing, sales, ops, science. Each domain has its own canonical chart vocabulary. We capture the distinctions, not just averages across them.
04
Commercial-grade evaluation infrastructure
Every label is documented, versioned, and traceable. Buyers can audit the dataset, evaluate model performance against it, and re-license updates.
Get in touch

Building AI that touches data visualization?

We license data, ship critique models, and partner with model labs and product companies that need to make AI better at charts. Talk to us.