ML Developer for the Team Automating Assessor Annotations

Yandex Crowd is a large-scale infrastructural service of Yandex. We use crowdsourcing to expand business processes: annotate data, moderate content, conduct field research and testing. We also create customer service and telemarketing, and handle localization and documentation.

Our team aims to automate assessor annotations. We are looking for an experienced ML specialist to join the team who will research, develop, pilot, and analyze. We work with generative models — LLMs and VLMs, develop co-pilots for assessors and annotation managers, and implement methods to improve quality and optimize annotations using models.

What tasks await you

Implementing generative models You will implement language and vision transformers, embeddings, optimize model inference, improve their quality, build pipelines, and develop microservices for integration into our platforms.

Researching SOTA approaches You will be tasked with finding optimal approaches to solve product tasks, researching the applicability of new approaches and SOTA architectures in the product, comparing internal and external models using benchmarks, and measuring their performance.

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

We expect that you

• Have worked with NLP, CV, STT
• Know the basics of DL and transformers
• Know Python well, have experience with production development, know the main principles of MLOps
• Want to become an expert who will independently develop and maintain a number of models and products
• Are ready to mentor junior colleagues, promote ideas, participate in hiring

Will be a plus

• Have optimized model inference (quantization, distillation, etc.)
• Have worked with agent frameworks