root@kernel:~$ cat /proc/identityI build research-shaped systems: agent harnesses with auditable authority boundaries, online judges that understand teaching data, and memory architectures that try to make long-context reasoning cheaper, sharper, and more falsifiable.
My favorite work lives where a mathematical claim has to become a running system — proofs, tests, architecture diagrams, telemetry, and a repo another engineer can clone, compile, and reproduce.
|
An auditable runtime for agent execution, memory, permissions, capabilities, and governed tool use. Git-like control plane between a user, their devices, their data, and autonomous agents. |
Multi-tenant competitive programming platform for educational institutions. Secure sandboxed execution with cgroups + seccomp, real-time WebSocket updates, LLM-assisted analysis, and 6-level RBAC. |
PyTorch research scaffold decoupling static memory lookup from dynamic reasoning. RetNet recurrence, Block Attention Residuals, and hashed N-gram Engram lookup — backed by 42 formal proofs. |
|
|
|
|
cachoidxx@gmail.com · github.com/XXY-CH · DOI: 10.5281/zenodo.20041183


