ML Developer for the Search Blender Team

Modern search results are not just a list of links, but a smart ecosystem of thematic blocks that helps users find answers faster and more accurately. The Blender team is responsible for ensuring that search models correctly interpret user intentions: should a weather block be shown for the query "pagoda"? and should a video be displayed for the query "14mm wrench"? To achieve this, we combine online and offline signals, train powerful ML systems, and make search smarter.

You will create ML models that directly impact millions of users and work on tasks where science and production are closely intertwined.

What tasks await you

Model Training You will train models to deploy them into high-load production.

Distilling LLMs into Compact BERT Models You will distill LLMs into compact BERT models for fast and efficient inference when selecting thematic blocks.

Developing a Flexible Architecture You need to develop a flexible architecture that will provide a unified foundation for dozens of blocks and allow for quick updates to the display logic of individual blocks.

Researching Generative Approaches You will conduct experiments with LLM + RL to build search results from scratch.

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

We expect that you

• Have experience with LLM or BERT-like models, as well as with RL
• Are proficient in Python and work with SQL
• Are passionate about ML, analytics, and integrating models into real services
• Have a good understanding of algorithms and data structures