The Hook
“For millennia, humanity advanced through billions of different minds working together.”
Problem Agitation
“So why are we scaling AI by just training one giant model?”
Product Reveal
“That's why we built Kardashev-0.7, the world's first trained swarm, made of 32 distinct models.”
Product Reveal
“I'm Ed. And I'm Cat. We're co-founders at Banbury Road. Our team met as top of the class undergrads at Oxford and left PhD at Yale and UChicago to scale intelligence horizontally.”
Feature Teaser
“With our technique, RL for population scaling, or RLPS, we trained 32 models to organically develop complementary capabilities.”
Feature Teaser
“Giving frontier performances at the level of GPT-5.5 for 1/50th of the cost.”
Feature Teaser
“For example, we solved the 70-year-old Pierce-Birkhoff conjecture in algebraic geometry for under $500.”
Feature Teaser
“Capability so far has scaled by model size, but we are scaling them by model count.”
Call to Action
“32 models is just the beginning. Be among the first to try Kardashev for yourself.”
Synthesizable in Remotion · key components of the cosmic-swarm-benchmark template.
Atmospheric curvature of Earth against deep space with soft orange sunset gradient along the rim.
Render SVG arc masked over dark radial gradient (#0b101c) with blur-filtered orange-to-blue linear stroke.
4x8 matrix of uniform hexagon wireframes morphing into distinct asymmetrical polygons.
Map 32 SVG polygon nodes using interpolate to transform vertex offsets from uniform radius to unique per-index values.
Horizontal or vertical bar animating from zero to score with highlighted first bar and text callout.
Use spring interpolation on scaleY and opacity, rendering comparative competitor bars in white/translucent and Kardashev in gold.
Purpose. Challenge monolithic AI by referencing human collective intelligence.
Execution. Founder talking head intercut with archival historical achievements.
Purpose. Unveil Kardashev-0.7 and founder technical pedigree.
Execution. Cosmic planetary title card leading back to founder presentation.
Purpose. Establish benchmark parity and 1/50th inference cost superiority.
Execution. Crisp multi-bar benchmark graphics and logarithmic scaling laws.
Purpose. Drive immediate developer and researcher trials.
Execution. Clean URL display against atmospheric horizon.
Objective
Launch of a distributed micro-agent database engineHow the recipe adapts
Swarm grid illustrates 64 distributed database shards morphing into coordinated pipelines, with benchmark bars comparing query latency against monolithic databases.Scene durations (9 scenes, 66s)
Key Remotion components
Introducing Griffin, the first AI to pass the video Turing test
The most emotive and controllable TTS model
We just released QVAC TranscribePsy, a speech-to-text model small enough to run inside a phone app. It covers 25 European languages on a laptop and 8 on a phone, and on an iPhone 17 Pro Max's Ne