ML Developer at Crypta

Crypta is an internal service that builds anonymized profiles of hundreds of millions of users based on their online behavior. Crypta's data is used by many Yandex services for personalization, including for selecting the most relevant ads.

Cross-device stitching is a set of solutions that allows linking different browsers and devices of a single user, as well as finding connections between users. Stitching helps Yandex account for a user's behavior history across different devices to improve personalization. To ensure stitching is complete and accurate, we actively use machine learning.

What tasks await you

Analysis of graph data and pattern identification You will dive into analyzing graphs linking billions of device and user identifiers. Your goal is to identify hidden behavioral patterns and regularities that will help to more accurately understand which devices and accounts belong to one person or are connected to each other.

Training and deployment of ML models for stitching A key part of the work is developing, training, and improving ML models (from classical algorithms to modern neural networks). You will improve the quality of stitching to ensure the most complete and accurate picture of each Yandex user's behavior.

Deploying ML into high-load services You will deploy your ML solutions into real production services that process colossal volumes of data - millions of requests per second, and optimize them. The task is to ensure the stability, speed, and efficiency of these systems so that the service does not choke under load.

Improving key business metrics for personalization Your work will be directly related to improving the quality of personalization, especially in Yandex's advertising services. You will constantly work on improving the key metrics of the Stitching service, which will ultimately lead to more relevant ads for users and increased company revenue.

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

We expect you to

• Have a deep understanding of ML: confidently apply classical methods and neural networks
• Be proficient in Python (Pandas, NumPy, CatBoost, Scikit-learn, PyTorch), SQL (optimizing heavy queries), C++ (for high-load components)
• Work with Big Data: have experience processing large distributed data

It will be a plus if you

• Have worked with graph data
• Understand web technologies: the nature of cookies, user agents, tracking mechanisms