Build intelligent systems we can understand all the way down.
Antshiv Robotics develops auditable CPU AI, control and simulation software, and the infrastructure needed to preserve how those systems were built and tested.
The long-term ambition is autonomous physical systems. The present work is more disciplined: harden the mathematics, kernels, runtimes, simulations, evidence, and hardware paths required to earn that capability.
Understand
Trace systems from equations and numerical contracts to executable kernels.
Measure
Use parity, profiling, and reproducible tests rather than unsupported claims.
Build
Connect CPU AI, simulation, control, and engineering infrastructure.
Share
Publish the process so others can inspect, reproduce, and contribute.
Intelligence is useful only when the surrounding system can be trusted.
A model, controller, robot, website, and experiment record are often treated as separate products. In practice they form one engineering loop: observe, compute, decide, act, measure, and retain the evidence.
AI should run on hardware people can own
CKE investigates how far CPUs can carry modern inference, bounded training, and eventually distributed execution without hiding the work behind a remote service.
Autonomy must be earned layer by layer
Dynamics, estimation, control, simulation, timing, embedded execution, and field validation must agree before a system deserves an autonomous label.
Engineering knowledge should survive the experiment
Antsand and ShivasNotes preserve data, decisions, failures, diagrams, and explanations so progress is not reduced to a final benchmark or product claim.
Three programmes at different stages of maturity
The programmes share an engineering method, but not the same evidence level.
C-Kernel-Engine
An open-source Linux CPU runtime and kernel compiler for transformer language, vision, and audio models, hardened through pinned numerical references and nightly tests.
Flight-control systems
Rigid-body mathematics, sensor models, state estimation, control, and deterministic simulation developed toward embedded and physical validation.
Antsand
Structured Databoards, federated websites, content pipelines, and evidence infrastructure used to operate and explain the engineering work.
The method is part of the product
The company is being built around engineering practices that make ambitious work inspectable instead of merely persuasive.
Start from first principles
Write down the equation, physical assumption, data contract, numerical reference, and intended behavior before optimizing implementation.
Fail closed
Missing kernels, unsupported routes, invalid shapes, and broken contracts should stop execution rather than silently substitute an unknown path.
Measure the active constraint
Use profilers, intermediate tensors, hardware counters, controlled experiments, and compiler matrices instead of optimizing from intuition alone.
Separate evidence from ambition
Measured, demonstrated, in progress, and planned are different states. Product language should preserve those distinctions.
Keep the record
Commits, pull requests, tests, technical articles, diagrams, and experiment lineage retain why a system changed and what the change actually proved.
Anthony Shivakumar
Anthony founded Antshiv Robotics to connect several bodies of work that had been developing separately: native systems software, CPU AI, mathematical modeling, robotics, data infrastructure, and technical education.
He currently directs the engineering, reviews the generated and contributed work, operates the testing and publishing systems, and documents the questions that remain unresolved. The company is deliberately honest about being early, small, and still building its strongest capabilities.
C-Kernel-Engine
CPU AI runtime and compiler
Antsand
Data, publishing, and deployment
ShivasNotes
Derivations, investigations, and evidence
Bring a concrete systems problem.
The best next conversation starts with a model, kernel, controller, simulation, hardware platform, experiment, or teaching problem that can be made reproducible.
Antshiv Robotics builds intelligent systems we can understand all the way down, from mathematical kernels and CPU runtimes to simulation, flight control, and autonomous machines.