Search Platform Quality Group Manager (Nova)

We are building a platform that will enable dozens of Yandex search verticals to develop faster. Today, the company has many vertical searches: products, travel, real estate, content, services. Each of them has its own ranking logic, its own quality metrics, its own interfaces for integration with products like Alice.

We want to create a unified search platform from these, where teams will be able to declaratively describe relevance requirements and receive ready-made ML search pipelines.

What tasks await you

Leading an ML team You will manage a team of engineers working on machine learning and search algorithms, define the technical strategy, prioritize tasks, and oversee the full development cycle — from model research to their deployment and monitoring in production.

Development and implementation of a DSL for describing search relevance You will design and implement a DSL for describing the logic of search result relevance and flexible configuration of ranking rules, factor weights, and filtering conditions without changing the system code, accelerate search quality experiments, and simplify algorithm management.

Using LLM as a compiler of ranking logic and for generating training data You will develop an approach where LLM acts as a compiler of ranking logic: converting declarative rules and task descriptions into ranking code and model configurations. Additionally, you will need to use LLM for generating synthetic training data and automatic labeling.

Building a pipeline for distilling ranking models for production latency You will create an ML distillation pipeline that allows transferring knowledge from computationally expensive ranking models into lightweight models for online inference, ensuring production-level latency without significant loss in search quality.

Building a metrics system for different types of search You will need to create a metrics system for various types of search tasks, define standardized metrics, calculation and reporting pipelines, which will allow teams to compare models and algorithm changes on a unified quality scale.

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

We expect you to

• Have managed an ML team
• Have launched ML systems into production
• Understand search architecture: retrieval, ranking, evaluation
• Know how to design platform solutions for multiple teams
• Can make architectural decisions under uncertainty

Will be a plus

• Have worked with search or recommendations
• Have worked with LLM systems and agents
• Have built ML platforms