Bounded Concurrency in Go: Admission Control, Cancellation, and Shutdown
A small admission gate makes overload and shutdown explicit. Follow the lifecycle from acquiring a slot to proving that every worker has finished.
Engineering notes
Akiwayne / Engineering notes
Go · Python · LLMSystems,
from the
inside out.
Go backends. Python engineering. The mechanics of model inference.
A personal notebook about how things work—and what happens when they don’t.
Independent notes / Reproducible experimentsScroll to explore ↓
01Selected writing
All articles ↗Structured Concurrency for Model Workloads in Python
Bound the queue and the number of tasks, preserve result identity, and make failures cancel the work they own. A standard-library asyncio lab.
Causal Attention from First Principles: Shapes, Masks, and Numerical Checks
Implement causal attention on CPU, trace every tensor dimension, and test outputs and gradients against PyTorch before reasoning about performance.
02From the workbench
Projects ↗Agent engineering
pi-go
Exploring the internals of a terminal coding agent through a Go rewrite: tools, sessions, and explicit failure handling.
Executable explanations
Engineering Labs
Six small experiments connecting backend reliability, Python concurrency, and the mechanics of model inference.
03Recent notes
Always a work in progressBounded Concurrency in Go: Admission Control, Cancellation, and Shutdown
A small admission gate makes overload and shutdown explicit. Follow the lifecycle from acquiring a slot to proving that every worker has finished.
Inside a Go Agent Loop: Tool Calls, State Transitions, and Failure Recovery
An agent is a state machine around unreliable boundaries. Build a deterministic loop that makes tool identity, errors, cancellation, and termination inspectable.
Structured Concurrency for Model Workloads in Python
Bound the queue and the number of tasks, preserve result identity, and make failures cancel the work they own. A standard-library asyncio lab.
Measuring Streaming Systems Without Fooling Yourself
Separate first-event latency, inter-event gaps, completion time, and failures. Build a small measurement tool whose arithmetic can be checked against known timelines.
Good systems start with a conversation.
Working on backend reliability or model-serving infrastructure?