ML Engineer RecSys at Market

The Yandex Market recommendations team creates personalized product recommendations on the homepage, selects similar and complementary products on product cards, and personalizes search. We are looking for middle/senior ML engineers for two tracks: neural network-based recommendations and a more classic RecSys + discovery.

What tasks await you

Neural network-based recommendations You will add new input data for neural networks: search queries, types of actions (sharing a product link), user features, etc. You will need to experiment with product embeddings (what to use from text descriptions and characteristics, how to account for product images) and with architecture (adding a separate head for a new target, changing the loss function, etc.).

You will also adapt models for scenarios: recommendations on the homepage, analogous products (accounting for the context of the main product), search (accounting for the context of the search query). Additionally, you will need to analyze data and find growth points in ML models through new targets, input data, or features.

Classic RecSys and discovery You will experiment with the target of ranking models. How to account for product unit economics? How to balance the user's current interest with new discovery categories? You will need to work on brand discovery (for example, how to show new fashion brands to clothing enthusiasts) and lifelong recommendations: answering the question of how to remember a user's relevant interests from the distant past. Example: bought a snowboard 1.5 years ago (interest/hobby - snowboarding) → we recommend snowboard boots or a ski helmet now.

You will also develop models for selecting analogous products: collaborative filtering, statistical co-occurrence (PMI, Swing...), BERT, etc. Furthermore, you will need to analyze data and find growth points in ML models through new targets, features, or business logic.

We expect you to

• Have worked in ML/DL for more than a year
• Be familiar with classic ML: CatBoost, ML metrics, methods of feature generation
• Be proficient in Python and SQL
• Know how to conduct A/B testing
• Have knowledge of NLP, BERT-like models

It will be a plus

• Have worked on search or recommendation tasks
• Have worked with LLMs
• Are familiar with neural network recommendation models: SASRec, HSTU