Project Alpine
A control plane for measuring and operating local AI inference, with a desktop workspace.
Control-plane & desktop development
The problem
Running a model locally involves more than starting a process. Different profiles need comparable measurements, reproducible evidence and a way to return to a known state when a change fails.
Engineering decision
I keep measurement separate from a claim of readiness. Qualification depends on identity-bound evidence and explicit gates; a successful launch alone does not mark a configuration as production-ready. Rollback and process identity are part of the design.
Workflow overview
Live inference and full production qualification have not been demonstrated in this portfolio. The workflow diagram is explanatory, not a live application.
Workflow
- Measure a baseline
- Compare a candidate profile
- Qualify or roll back
An explanation of the workflow, not an interactive application.
How it works
Alpine coordinates discovery, measurement, profile selection, qualification and session lifecycle. Rust owns the control plane, SQLite records evidence, and the desktop interface brings the workflows into one workspace.
Collaboration & credits
Inference is provided by llama.cpp and its runtime. Alpine is the orchestration and evidence layer, not a new inference engine. The desktop uses React and Tauri.
Verification details
The desktop type check and 34 interface/task tests passed in the portfolio verification run. The source contains the control-plane workflows described here.
- Evidence date
- Not recorded
- Source revision
5e28a3011b8b2fce5112efc0366ec6a4c8206d15- Working tree
- modified
The recorded source had working-tree changes. The exact historical patch was not retained; this revision alone cannot reproduce that demonstration.