🚀 When Algorithms Understand People: How "Leisurely" and "Lively" Routes Appeared in Maps

Our team at Yandex Maps works every day to make city travel not just fast, but truly comfortable. Especially for those traveling by foot, scooter, or bicycle.

Recently, my colleague Anton Ovchinkin, team lead of the pedestrian and transport navigation group, shared details on Habr about a major team achievement. Now, alternative routes in Maps have character: you can choose not just a "path 5 minutes longer," but consciously go via a "Leisurely" route through a park or a "Lively" route along well-lit streets.

The hardest part was teaching the model to understand subjective human preferences. To do this, the team used an interesting approach: they employed LLMs for data labeling and trained a lightweight CatBoost model that operates in real-time. Read Anton's article to find out how they managed to stay within strict timeframes and scale the solution to 17 cities. This is a great example of how complex ML tasks are solved for millions of users.

This very team is the one behind building thousands of routes per second, developing Yandex MapKit, and providing navigation for Yandex Eda and Lavka services. And right now, we are looking for a C++ Developer to join their ranks.

We are looking for someone who:

• Is proficient in C++ and enjoys complex algorithmic tasks.
• Is ready to work with high-load backend services and big data.
• Shares our interest in a product that millions see and use.

If you want to develop technologies that truly change urban navigation, then you belong with us.

Company Type