ML Developer for the Advanced Advertising Products Development Team

ML Developer for the Advanced Advertising Products Development Team

Advertising is one of Yandex's largest and most technologically complex products. It is important for us to ensure the high-quality performance of our advertising: showing users ads that are most likely to interest them. We develop neural networks that are applied in real-time for ad selection and ranking, as well as for solving generative scenarios.

Today, through Yandex Direct, ads can be placed on many platforms: in Search, on partner sites (RSYa), in Telegram channels and on other bloggers' platforms, and on retailer websites (Retail Media). Our team improves the effectiveness of these platforms, which is why we solve interesting tasks related to ML, product, and infrastructure challenges on a daily basis.

What tasks await you

Development of new ad targeting algorithms In addition to the ad itself, the advertiser provides us with targeting information — to whom they want to show their ads (target audience, description of search queries, description of bloggers). Previously, the task of ad matching was solved algorithmically, but now we actively use neural networks. One of our global goals is to increase the convenience of setting up advertising campaigns, for which we are developing and implementing new interaction tools.

You will work with editorial, assessor, and model annotations, process big data, train heavy ML models, and convert them into lightweight versions for real-time use.

Improving ad selection processes considering the specifics of creatives Ads can look different. Yandex provides a convenient search by themes. Ads for such formats contain much more useful information for the user (for example, for financial topics, these are banking characteristics of deposits and cards; for product topics — the cost and characteristics of the product). For advertising on bloggers' platforms, long texts with several pictures and videos are used.

You will experiment with and develop the effectiveness of matching various ad creatives: work with text, images, numerical, audience, and statistical features in ad selection and ranking models.

Training generative models In our products, there are various scenarios where our partners can use Yandex's generative solutions, built on top of the Alice AI model family. Our generative models help bloggers write promotional posts in the style of their blog, correct errors in posts, create new posts based on descriptions, and summarize them into digests.

You will improve the base quality of our generative models, experiment with using external information and user feedback, and implement new generation approaches into different interfaces and dashboards.

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

We expect that you

• Know Python and C++
• Have a basic understanding of algorithms and data structures
• Have experience training models in the field of NLP: BERT, GPT, and others
• Are familiar with recommender system architectures
• Are familiar with A/B testing and can conduct basic analytics on the impact of your implementations