ML Developer for the Visual Yandex Market Quality Team

Our product is the visual ranking of Yandex Market. We enhance the user shopping experience using visual content, helping users find beautiful, visually appealing products online. Our team tackles a very complex task: to collect products and their images from across the internet and present the user with the most relevant offers for their query.

What tasks await you

Merging offers from the internet into a single entity (deduplication) One of the main focuses this year is to improve the user experience when selecting products in visual categories. We are solving an extremely difficult task: to gather product data from various sources and combine it into a single product card, providing the user with a convenient interface for selection.

Training ranking models We actively work on ranking products and images. Developing ranking involves both working on numerous factors used by the models and experimenting with approaches to their training.

Improving personalization in product and image ranking We are building personalized ranking and trying to provide the user with exactly the product they want to buy. For this, we train many different models: from simple query/document classifiers to ranking DL models that take into account user history and profile.

Improving the accuracy of document-query matching One of the key tracks is increasing document relevance to the query. We actively work with assessor relevance labels, train large LLM/VLM models based on them, and distill complex production solutions from them.

More about ML at Yandex — in the channel Yandex for ML

We expect you

• Have worked with PyTorch or are ready to quickly dive into it
• Are proficient in Python and C++ at a level sufficient to implement your ideas
• Have experience in one of the areas: RecSys or Computer Vision
• Like to take initiative and propose your own solutions to product problems

Will be a plus

• Have worked with data labeling, for example in Yandex Toloka or Yandex Tasks
• Have worked with modern recommendation systems and deep Doc2Vec models
• Have developed solutions for services with critical runtime requirements and understand the specifics of training and optimizing models for them
• Understand the specifics of product search