OpenJ / JournalBuild notes, experiments, useful failures
Notes from the lab
Follow the work while it is still moving.
The journal is where implementation details become explanations: what was tried, what broke, what measured better, and what became reusable.
From Deliverance to Reckoning
Giving the inference engine a home inside the broader OpenJ project.
Making tokenizers useful from Groovy
Why a Python-shell-like workflow belongs in the JVM toolkit.
What happens beneath the model API
Tensors, quantization, caching, and the path from prompt to output.
When a benchmark becomes a design document
Using measurement to decide which runtime details deserve attention.