AI & Scientific Infrastructure
Collapsing the 40-year gap between discovery and application. Where imagination and desire replace specialized training as the true competencies of the inventor.
The Fourth Domain
The three pillars of Atumics — Omniscience, Omnipotence, Omnipresence — are research ambitions. But ambitions need tools. The AI and Scientific Infrastructure programme builds the tools that make the pillars possible: autonomous research agents, generative design systems, synthesis-on-demand capabilities, and the vision of General Solution Synthesis.
The generational barrier between a scientific breakthrough and its engineering application has historically been 20 to 40 years. AI systems capable of synthesizing knowledge across every field simultaneously can collapse that barrier. The inventor of the future needs imagination and desire — the AI provides the synthesis.
Atum Scientist
Autonomous Research Agent
An AI system architecture designed to conduct scientific research with minimal human intervention. Atum Scientist ingests experimental data, builds internal representations of natural phenomena, generates hypotheses, designs experiments to test them, and integrates results back into its world model.
Unlike conventional ML systems that require human-designed equations and features, Atum Scientist learns directly from data — representation-less modeling. It discovers structure in nature that humans may never have formalized, then translates its internal representations into human-readable forms: mathematical, graphical, and textual.
Deployed internally. Under evaluation for licensed deployment at CERN and ESS (European Spallation Source).
Key Capabilities
- • Representation-less modeling — learns directly from experimental data
- • Automatic hypothesis generation and falsification
- • Experiment design and, via Gluino, automated execution
- • Human-readable translation of internal representations
- • Cross-domain knowledge synthesis spanning physics, chemistry, biology, and engineering
Atum Engineer
Generative Design System
The applied counterpart to Atum Scientist. Where the Scientist discovers knowledge, the Engineer applies it. Given a mission or product requirement expressed in natural language, Atum Engineer searches the patent literature and materials science databases, synthesizes candidate designs through AI-driven concept generation, validates designs through simulation, and produces complete manufacturing specifications.
A researcher can describe a need — "an antenna that operates at 300 GHz with 60 dB gain in a 2mm package" — and Atum Engineer will propose multiple candidate architectures, evaluate their trade-offs, simulate their performance, and output fabrication-ready designs through Gluino.
Deployed internally and available to strategic partners for next-generation component design.
Synthesis-on-Demand
AI-Driven General Solution Discovery
An internal research capability, increasingly offered as a service. Given a set of performance requirements — "an energy storage medium with 10x the density of lithium-ion and no flammable components" — the system searches the full space of known physics, chemistry, and materials science to propose novel candidate solutions.
These include structures that no human researcher has considered, because they require knowledge from multiple disciplines that rarely interact. Each candidate is validated through simulation, then routed to laboratory verification through Gluino.
The origin of this capability traces to the 2024 informal research report "General Solution Synthesis" by Morten Skærsø, which argued that the combinatorial space of physical possibility vastly exceeds what any human specialist can explore, and that AI systems unconstrained by disciplinary boundaries would discover solutions invisible to conventional research.
The Vision: General Solution Synthesis
The end-state of our AI programme is a system that can take any well-posed engineering challenge — expressed in the language of desired outcomes — and produce a complete, validated, manufacturable solution. Not through lookup or interpolation, but through genuine synthesis: combining principles from physics, chemistry, biology, and engineering in ways that no single human expert could.
This is not artificial general intelligence in the philosophical sense. It is artificial general engineering — the automation of the inventive process itself. We believe this is achievable within our generation, and we are building it one capability at a time.