Projects

the exactly-once store

ProcessMesh

Durable execution for side effects that must happen exactly once.

A distributed backend where every side effect is idempotent, every retry is safe, and a crashed worker never leaves a half-finished job behind.

the latency path

ChronosBook

A C++ order book tuned for latency, memory layout, and cache behavior.

Low-level performance work: contiguous memory, cache-line-aware structures, and measured tail latency under sustained load.

the scheduler and the workers

Aster

A fault-tolerant distributed task runtime built from scratch in Python.

Schedules dependency-aware workloads across local and remote workers. Survives worker failure through heartbeats, retries, resource reservations, backpressure, and stale-result protection.

AtlasSearch

TODO: one-line claim.

TODO: two-sentence summary.

XDP/eBPF L4 Load Balancer

Kernel-level L4 load balancing in eBPF. In progress.

TODO: two-sentence summary. Replaces ChronosBook as flagship #2 for infrastructure roles when it ships.

Orbrick

TODO: identity / auth one-liner.

TODO: two-sentence summary.