Data-driven
scientific visualisation
People are wired to think visually — a table of raw data doesn’t communicate on its own, no matter how important the underlying research is.
Traditional data visualization
Accurate, but built for scientists, not for anyone else — dry, technical, and easy to tune out.
Traditional motion graphics
Visually engaging, but disconnected from real data. Hand-animated to look a certain way, not driven by an actual dataset — so however good they look, they can’t represent real research at all.
For a company trying to raise funding, that gap has direct consequences: convincing investors depends on communicating a genuinely technical process in a way non-specialists find compelling, and neither existing approach does that alone.
Process
01 · Import dataset — the AI denoising data coming directly from Orbital’s system
02 · Import periodic table — reference so each element gets a consistent, physically-accurate size and colour
03 · Match databases — cross-reference dataset elements against the periodic table, so e.g. hydrogen renders at correct relative size and colour every time
04 · Remap timestep to frame — maps the dataset’s per-particle frame number directly onto Houdini’s timeline, so playback is accurate without manual timing
05 · Create intersecting lines — optionally visualizes connections between elements where relevant to the story
06 · Create shaders — shades each element based on its matched colour data
07 · Add background, create lights, render
Outcomes
Funded
The resulting animations were used as part of Orbital’s investor materials for a funding round that closed successfully
Repeatable
A repeatable system for visualizing Orbital’s AI denoising process — any new dataset from their pipeline could be run through it without rebuilding the process by hand
Challenges and future direction
Manual lighting & environments
Backgrounds and lighting are currently set manually per project. Automating that — generating appropriate environments and lighting conditions based on the data or context — is a clear next step.
Manual title/text layer
The final deliverable was a video with animated text and titles layered on top of the rendered data. That layer is currently manual too, but doesn’t need to be — using Cavalry, title and text animation could be driven directly from a Google Sheet or another dataset, closing the gap between raw data and finished video entirely.