Writing
Write-ups of projects I built and what I measured. For background on me, see about.
projects
Overviews of each project.
- SequenceAn unfamiliar system is supposed to be readable if you put the hours in. In practice the limit is that prose about an architecture is a list you reassemble in your head every time you reason about it. This hands the structure back as something you can point at.
- ML HarnessTraining a model is supposed to be the hard part. In practice the training is the step with the most written about it and the least left to decide, the difficulty is what to train, on what data, and whether the machine in front of you can hold it. This is an attempt to make those decisions walkable on one consumer GPU.
deep dives
Longer write-ups on one topic each.
- Building an internal knowledge baseWhy an enterprise should own its knowledge base, and how live usage keeps it current.
- Sequence: architecture as a graphFeeding agents deterministic, structured architecture so they build the right thing instead of drifting from intent.
- A self-optimization sandbox for AI systemsA closed loop that improves our AI copilots and proves every gain against an independent judge, and where it points next.