Pienso Metamorph
Automated deep-learning training that informs and delights
- UI
- UX
- AI/ML
- 0->1
Model-training in progress. Watch the full-length video here.
Metamorph is a deep learning training interface that tracks model quality in real time so users can spot issues early and take corrective action.
Team
Design Will Crum
Development Jack Murphy, Brian Cort, Felipe Balduino Cassar
Background
We wanted to enable our users to turn labeled training data into accurate deep learning models for text classification.
Objective
Provide a descriptive, delightful training UI that lets users observe model quality in real time — and spot issues early.
Process
0-1 stand-up of a new feature MVP. I sketched, wireframed, mocked up, and refined in build with the dev team. Briefed in Oct 2021, MVP built by EOY, with further refinement in 2022.
Challenges
- Hands-off UX – Deep learning training is an automated, wait-and-see process.
- Many metrics – Performance stats are available, but technical. Which would actually inform users?
- Delight them – Could this slow, dry process be visually appealing?
Background
- Capability, delivered – Users could now convert their labeled data into more accurate deep learning models
- Actionable info – Our real-time metrics let users immediately spot if model was failing, and why.
- Data viz delight – Users enjoyed watching the charts warp and wiggle