Language Intelligence
LLMs, reasoning, agents, structured intelligence.
Research-led deeptech · Frontier AI · Advanced computing
Engineering the Next Frontier
Arxigen builds its own AI software, starting with two products: Causal State Delta for faster AI agents, and Self-Correcting Gaussian Splatting for 3D scenes that repair themselves.
Why we exist
Most breakthroughs stall somewhere between the paper and the product. Arxigen exists to close that gap, carrying its own research through every stage below.
- 01 · RESEARCHCurrent stage Open questions Fundamental work in AI and computing, chosen for what it could make possible.
- 02 · TECHNOLOGYNext Proven methods Results made robust, measurable and reproducible outside the lab.
- 03 · PRODUCTNext Real problems Technology shaped around specific users, constraints and outcomes.
- 04 · PLATFORMNext Systems at scale Infrastructure that others can build on, reliably and for years.
Research areas
Where we work
Six research areas, each pursued with a path to deployment in view.
Multimodal AI
Language, vision, audio and cross-modal reasoning.
Generative Systems
Synthetic data, diffusion, world models and simulation.
Efficient AI
Inference acceleration, optimisation and model efficiency.
Perception & Spatial Intelligence
Vision, sensing, 3D/4D and scene understanding.
Scientific AI
AI for science, engineering and emerging technology.
Animations are illustrative. Figures shown are not measured results.
Products
What the research becomes
Two proprietary products, researched, built and owned by Arxigen. Each is designed to be measured, verified and released as software others can build on.
P.01 In research
Causal State Delta
AI agents that read only the changes that matter to their next move, instead of re-analysing the whole screen every frame. For agents that operate games, websites and 3D software: faster decisions, same accuracy.
How it works From R.04 · Efficient AI AI agents & interactive software
P.02 In research
Self-Correcting Gaussian Splatting
3D scenes that find their own geometric errors, work out why they happened, and repair only the broken region, then verify the fix. For digital twins, robotics mapping and 3D content: no full reconstruction reruns.
How it works From R.05 · Perception & Spatial 3D reconstruction & spatial computing
How we work
Principles
Evidence before ambition
We scale what we can measure. Claims are tested, benchmarks are honest, and negative results are kept.
Engineering is research
Making an idea work at scale is a discovery problem in its own right, and it gets the same rigour as the idea itself.
Built for the long horizon
We choose problems that will still matter in a decade, and build systems meant to last that long.
Founder
One founder. Frontier problems. No committees.
Founder & CEO
Bhuvaneswari Babu
Leads the research and engineering behind Arxigen’s two products, Causal State Delta and Self-Correcting Gaussian Splatting. A full-stack developer with hands-on experience integrating AI into applications, from interface to APIs and databases, and holds an Anthropic certificate in building with the Claude API.
Team size 1
Every hypothesis, experiment and line of code has one owner. Fewer meetings, faster iterations, a straight path from idea to working system.
HYPOTHESISEXPERIMENTWORKING SYSTEM
Questions
Frequently asked
What does Arxigen do?
Arxigen is a London-based deeptech company that builds its own AI software. It is developing two products: Causal State Delta, for faster AI agents, and Self-Correcting Gaussian Splatting, for 3D scenes that repair their own errors.
Is Arxigen a consultancy or agency?
No. Arxigen researches, builds and owns its own products. It does not take on client projects.
Where is Arxigen based?
Arxigen Ltd is registered in England and Wales (company number 17518725), with its registered office in London.
What is Causal State Delta?
Causal State Delta is an approach for AI agents that operate games, websites and 3D software. Instead of re-analysing the whole screen every frame, the agent reads only the changes that matter to its next move, aiming for faster decisions at the same accuracy.
What is Self-Correcting Gaussian Splatting?
Self-Correcting Gaussian Splatting is a 3D reconstruction method in which a scene finds its own geometric errors, works out why they happened, repairs only the broken region and then verifies the fix, with no full reconstruction rerun. It is aimed at digital twins, robotics mapping and 3D content.
Can I use the products today?
Not yet. Both products are in research. Ask for early access through the contact form and Arxigen will be in touch when they are ready.
How can I get early access or collaborate on research?
Use the contact form at the bottom of this page and choose “Early access” or “Research collaboration”.
Get in touch
Stay close to the work.
Arxigen builds its own products. Ask for early access when they are ready, propose a research collaboration, or send any other question.