About
Our team cares about protecting the environment by optimising generative AI to consume fewer resources. Our approach makes interface generation substantially more efficient, and our mission is to make it available to everyone who needs it.
Doing more with less
Bilboard is an agentic AI platform that builds analytical dashboards from natural language. What makes it different is where the model's effort goes. Most tools ask a large language model to write the interface code itself, which means paying for thousands of tokens of framework boilerplate on every request and regenerating the same scaffolding from scratch each time.
Bilboard takes a schema-driven approach instead. The model produces only a compact, constrained description of the dashboard, an intermediate representation drawn from a closed vocabulary of components. That description is validated against a schema before anything runs, and a deterministic rendering engine turns it into working components. The model decides what the dashboard should contain and the engine handles how it is built.
The efficiency gain follows directly from that separation. In an exploratory comparison on dashboard construction tasks, the schema-driven pipeline consumed roughly 6,000 tokens per task against roughly 17,000 for direct code generation, a reduction of about 65%. Fewer tokens mean less inference, and less inference means less computation, less electricity and less water drawn for data-centre cooling. The boilerplate is never generated, so its environmental cost is never paid.
Built for minds that think in shapes
Dense text is a barrier for a great many people. For those with ADHD, long paragraphs of figures demand sustained attention that is expensive to give and easy to lose. For autistic people, unstructured prose can be harder to parse than the same information given a clear, predictable visual form. In both cases the difficulty lies not with the data but with the way it is packaged.
Data visualisation changes the packaging. A trend line shows direction without asking the reader to hold a column of numbers in working memory, and a bar chart makes a comparison immediate rather than something to be reconstructed sentence by sentence. Colour, position and size carry meaning that would otherwise have to be read, held and compared, shifting the effort from sustained reading to a single glance.
Bilboard lowers the barrier to producing those visuals. Someone who wants to understand their data describes what they want in plain language and receives a dashboard with no code to write, no chart library to learn and no configuration screens to work through.
Why the output stays predictable
Three decisions do most of the work, and they are the reason the same request produces the same dashboard every time.
A closed vocabulary
The model chooses from a fixed set of components rather than inventing markup, so every dashboard is assembled from parts the renderer already understands.
Validated before it runs
Each description is checked against a schema first, so a malformed dashboard is caught at the boundary instead of failing in front of the person who asked for it.
The same request, the same layout
Rendering is deterministic, which matters a great deal for users who rely on consistent structure to navigate comfortably.
Analytical insight, available to everyone
The ambition is straightforward. Data analysis should not require a particular way of reading, and Bilboard is one step towards making analytical insight available to people whose minds work best with shapes rather than sentences.