Data Scientist

About the role

Every quiz on Free Online Quizzes is generated by a language model: seven questions, four options, a hint, in one of 38 languages, across 24 categories. That produces a lot of questions, and nobody has yet measured which of them are actually good. You would be the first person whose job that is.

You will work directly with the founder across the whole stack — from the raw generation logs and Firestore user documents to GA4 and PostHog — and your findings ship straight into the product, usually the same week. There is no data team to join; you are starting it.

What you will do

• Build the first quality model for generated questions: detect wrong answers, ambiguous options, leaked hints and questions that are trivially easy or unanswerable, and feed the signal back into the generation prompts.
• Calibrate the easy / medium / hard difficulty tiers against real answer rates so the labels mean the same thing in every category and language.
• Design the recommendation logic for what a player is offered after finishing a quiz, and measure it against the current random pick.
• Analyse retention and the leaderboard: who comes back, what a reset does to engagement, where the multiplayer funnel leaks.
• Set up the experiment framework in PostHog so product changes are read with a p-value rather than a hunch.
• Write up what you find in plain language for a non-specialist audience — the founder and, sometimes, the blog.

What we are looking for

• Three or more years doing applied data science or analytics on a live product, with results you can point to.
• Fluent Python and SQL; comfortable pulling data out of Firestore, MongoDB and a warehouse without waiting for someone to build a pipeline.
• Practical experience evaluating LLM output — grading, rubric design, or building classifiers over generated text.
• Sound statistics: you know when an A/B test is underpowered and say so.
• Clear written English. Most of your work reaches the rest of the team as a document.

Nice to have

• Experience with recommendation systems or bandits in a consumer product.
• A second language among the 38 we serve — question quality does not look the same in Vietnamese as it does in English.
• Familiarity with GA4 and PostHog.