AI Systems

The infrastructure, evaluation, and systems work behind machine learning models.

Inference engines, evaluation, model systems, agents, data pipelines, efficiency, and the tooling that makes AI systems usable.

Start here

You should be able to evaluate, serve, and debug a model as a system component rather than a chat interface.

Test your understanding — adaptive checks that estimate what you actually know, question by question.

Query, key, and value projections, scaled dot-product attention, masking, and multi-head structure.

Autoregressive decode, KV cache memory growth, batching, and the serving cost model.

Crapht OS: strong defaults for hard conditions
Curation plus doctrine plus community, standing on Arch. Strong defaults for hard conditions: offline-capable installs, potato PCs as a primary target, power-resilient updates, a local model as the default agent, and every part open to inspection and change. Nobody writes a kernel. Full vision: docs/vision/crapht-os.md in the Crapht repo.

No Notes yet.