The role
Lanternfish builds small assistants that run fully offline on older phones. Our models are judged on one thing above all: knowing when they don't know. You'll own how we measure that.
You'll
- Build and maintain our internal benchmarks, including refusal and uncertainty tests
- Write test sets with native speakers for the languages we support
- Turn messy user reports into reproducible evaluation cases
- Present results to the whole team every fortnight, plainly
You have
- Experience evaluating machine learning models or building test infrastructure
- Solid Python and data analysis skills
- Respect for qualitative failure analysis, not just leaderboard numbers
- Willingness to work from our Oslo office three days a week
Perks
- Relocation support
- Five weeks' holiday
- A team cabin trip every winter
