ML Developer for the Generative Responses Trigger Group

Alice in Search is a key Yandex product: generative responses at the top of search results are seen by tens of millions of users daily. Under the hood, there isn't just one LLM but a whole family of models with different sizes and properties: some respond with images, others solve math problems, suggest products, or generate concise answers. Our team determines which model to run for a specific query and when it's better to skip generation altogether (for example, if a user simply searches for "vk").

What Tasks Await You

Development of Intelligent Query Routing Models You will work on query classification, selecting the optimal model for generation, and predicting the quality improvement from using a specific model.

Working with Data and Noisy Signals You will collect and prepare datasets from logs and annotations, work with LLM-based annotation, and train models on incomplete and noisy data.

End-to-End Cycle: From Experiments to A/B Tests You will define targets, train and calibrate models, run offline evaluation, and conduct A/B experiments on live traffic with a large audience.

Engineering and Product Solutions at the Intersection of NLP, LLM, and Classical ML You will participate in selecting architectures, configuring pipelines, and discussing product decisions that impact the quality of generative responses for millions of users.

We Expect You To

• Have a solid understanding of modern NLP and have solved product tasks using it
• Know how and want to solve full-cycle product ML tasks: from data collection to production deployment
• Understand the principles of LLM operation and are interested in tasks related to routing and combining models