Device concept
Week 2: first detections
Week 3: the dashboard comes online
Week 4: reading the data
Week 4 was mostly reading the dataset before anything was trained: indexing every frame, and plotting how pain intensity rises and falls across a sequence. Pain turned out to build and hold rather than spike, which told us the sequence model was the right shape. Written up in the journal.
-
How pain intensity is distributed across all frames: almost all of them score zero. Written up in the journal.
-
What a patient reports against what their face shows: the two only agree moderately, even by a trained observer's reading. Written up in the journal.
Week 6: models trained, hardware arrives
-
Model training progress, Week 6. Written up in the journal.
Week 8: showcase day
-
The redesigned carer dashboard, versions A to F: four beds on the ward view, live alerts and the night's history at a glance. The bottom row is the version we built. Written up in the journal.
The charts from each week live in the journal too, where they have context. New weeks get added here as each one runs.