← Back to selected work

Product & engineering

Bringing learning across languages.

Translation quality, automation, and model tradeoffs in a programming education product.

My role
Overseas development group leadership
Organization
QUREO
Timeframe
Presented in 2024

Context

Programming lessons need more than literal translation to work in another language.

The challenge

Terminology, lesson structure, and on-screen context all affect whether translated material makes sense.

The approach

  1. Automate structural checks and screenshot capture with Cypress.
  2. Compare custom translation models and LLMs against the needs of the learning material.

Protect the learning experience

Translation is part of product delivery. Correct data structures alone do not establish whether a lesson works for a child; visual and editorial review remain important.

Choose for the task

Model quality, preparation cost, and ongoing usage cost all belong in the decision. The newest model is not automatically the right choice.

What changed

Automated checks and a more practical visual-review workflow supported translation quality.

What the evidence tells us

The public account describes workflow improvements. It does not quantify attributable revenue or total time saved.

Sources & further reading