ML Developer for Ecom Blender (Search)

A modern search results page is not just links to relevant websites, but also a set of diverse thematic blocks that help fulfill a user's query more accurately and quickly. For example, for the query "kitties," the search results could appropriately include both links to resources with photos of kitties and images and videos directly on the results page. This variety of elements helps users solve their tasks faster.

Blender is the technology that allows mixing documents and thematic blocks from various data sources in the search results and selecting the most relevant ones for the query. We are focused on developing e-commerce scenarios in Search — this is when a user is looking for products online. Besides links to stores, Yandex can aggregate the best offers, show detailed product cards, help inspire purchases of different items, and do much more. Blender must use ML models to form the most useful search results page for the user.

What tasks await you

Forming the best e-commerce search results You will solve complex ML tasks related to placing thematic blocks in the search results from idea to production implementation, optimize scenarios for selection and purchasing online, conduct online and offline experiments, and work with both classical ML and neural networks.

Implementing new product solutions User queries are constantly changing, new directions and types of information are emerging, and along with this, the search results page is evolving, forming different thematic blocks. You need to know how to display them optimally. To do this, it is necessary to improve model architectures, select the right data for training, test various hypotheses, and analyze the results.

Implementing LLM for building search results You will be forming search results using LLMs. It is necessary to determine the most useful blocks for answering queries, analyze user behavior, the content of thematic blocks, and improve the search system based on the obtained data. We don't just conduct research, but implement our solutions in one of the most high-load search systems.

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

We expect that you

• Understand the variety of modern LMs and have solved applied tasks using them
• Have a good understanding of the principles of classical ML and have applied it in practice
• Know Python and C++, both or separately, and are willing to learn the second one

Will be a plus

• Understand neural network architectures
• Are familiar with big data processing technologies