Three systems. One method of turning mathematics into evidence.
C-Kernel-Engine executes modern AI on CPUs. Flight-control systems connect estimation and control mathematics to physical machines. Antsand preserves the data, experiments, and explanations around both.
These programmes have different maturity levels. We label what is demonstrated, in progress, and planned instead of presenting a roadmap as a finished product.
C-Kernel-Engine
Demonstrated and actively hardened
Flight control
Mathematics and simulation in progress
Antsand
Operational content and data system
C-Kernel-Engine: AI from mathematical circuit to generated C
CKE is a publicly developed Linux CPU runtime and kernel compiler for transformer-based language, vision, and audio models. It treats a model as explicit circuits, numerical contracts, kernel choices, memory plans, and generated native execution. Its repository is public for inspection; a recognized top-level reuse license has not yet been declared.
Runnable model families
Qwen, Gemma, multimodal, recurrent-attention, and other model paths are brought up against pinned references rather than accepted because they produce plausible text.
Parity before performance
Operation-level comparison, first-divergence tools, zero-tolerance or declared-tolerance tests, compiler and ISA lanes, and nightly regression evidence define correctness.
Owned and accessible compute
The research asks how far commodity x86 and Arm CPUs can carry inference, bounded training, and eventually distributed execution when kernels and circuits are designed for them.
Not a universal performance claim
AVX2 is the most hardened lane. AVX-512, AMX BF16, additional models, training, and multi-node execution remain evidence-gated work.
Flight-control systems: equations before autonomy claims
The flight-control programme builds the mathematical and software path from rigid-body dynamics to estimation, control, simulation, embedded execution, and eventual physical validation.
Dynamics and kinematics
Reference frames, rotation matrices, quaternions, forces, moments, actuator response, and six-degree-of-freedom motion form the plant contract.
Sensors into state
Calibration, filtering, covariance, Kalman-family estimators, and deterministic sensor models turn noisy measurements into usable state estimates.
State into action
PID and later model-based controllers are evaluated against explicit stability, timing, saturation, and failure assumptions.
Simulation before flight
Equation tests and simulation precede hardware-in-the-loop, embedded targets, controlled flight, and any claim of autonomous operation.
Antsand: structured content, data, and engineering memory
Antsand is the application and publishing infrastructure behind the websites, Databoards, content pipelines, structured records, and federated deployments used across this work.
Databoards over page fragments
Pages are composed from structured sections, data contracts, reusable templates, and site-owned Sass instead of being trapped in isolated hand-written documents.
Research into public explanation
Git history, experiments, diagrams, blogs, carousels, and partner records form a traceable content pipeline rather than disconnected marketing output.
Federated sites
One system supports distinct websites and brands while preserving independent routes, content, styles, deployment state, and operational ownership.
Engineering evidence layer
The long-term role is to connect experiment lineage, measurements, technical decisions, and communication around AI and robotics systems.
Model or sensor → computation → action → evidence
The programmes converge when AI and control software must execute correctly on real hardware and leave enough evidence for another engineer to inspect the result.
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.