NBT evaluation platform

An internal web platform for model and data management, evaluation runs, result tracking and data sharing, designed with interfaces for bad-case analysis, returned-data management and map-building visualization.

Tencent Maps 2021 – 2022 Designer and developer

1place for models, data, evaluations and results
  • Evaluation
  • Web development
  • MLOps
On this page

The problem

Teams needed one shared place to manage models and data, run algorithm evaluations, keep results and share data, instead of each person tracking these separately.

Design

The platform puts models, data, evaluation jobs and results behind one web interface, with an API so other tools can plug in.

Evaluation platform architecture
Platform architecture
  • Model and data management in one place.
  • Algorithm evaluation run through the platform, so results are produced the same way.
  • Result management and data sharing across the team.
  • Interfaces kept open for bad-case analysis, returned-data management and map-building result display.
Evaluation platform interface design
Interface design

The product

Evaluation platform, version 1
Version 1
Evaluation platform, version 2
Version 2

Lessons

  • Comparability is a product feature. A standard evaluation path matters more than any single metric.
  • Design the interfaces before the features. Leaving room for bad-case analysis and data return keeps the platform extensible.