DeadBench
Tools for lab work
The shelf is what we publish. The bench is where it gets made — notebooks for building models, a scheduler that runs them against the apps, and the data behind both. Private, by invitation, and quiet.
Tools
NotebooksJupyterHub
Where models get built. A shared Python stack — the same one the scheduler runs — so what works in a notebook works in a job.
Open →
Schedulescron + one-shot containers
Regular jobs for the apps — feed scoring, terrain models, retraining as data comes in. Each run is a fresh container from the notebook image; nothing stays resident between runs.
Logs land on the bench
DataPostgreSQL
One shared instance, one database per app. The notebooks and the jobs read it over the internal network. The internet does not.
Reachable from the bench only
How work moves
- 01 Prototype Explore the data and build the model in a notebook. Break things where it is cheap to.
-
02
Promote
Move the working code into a script in the shared
jobs/folder. Same image, same libraries, no surprises. - 03 Schedule One line of cron runs it on a cadence as a throwaway container and writes the result where the app reads it. Logs kept, nothing left running.
Access
The bench is private. Accounts are by invitation — if you need one, or something here is broken, write us.
doctor@deadkneelabs.ca