ML Developer for the Agent Solutions Development Team

Yandex Code Assistant is a product based on large language models (LLMs) designed to enhance development efficiency. Here's what we're working on:

• A YQL agent-assistant that finds and fixes errors in queries
• A Nirvana agent-assistant (a computational platform for creating and automating computational processes and data processing) — it will explain why a process crashed and how to fix it, and help find specific information
• A code agent — helps write code, explain it, search internal documentation, and more
• Code Completion — smart code autocompletion
• Pull request description generation

What tasks await you

Design and development of a high-load multi-agent service Thousands of colleagues solve work tasks using the agents we create. Our service must withstand load even during peak hours, guaranteeing stability and response speed. You will need to optimize GPU inference performance, accelerate streaming, write highly efficient and maintainable code, quickly diagnose errors, and stabilize the service.

Development of the multi-agent system We are integrated into many components of Yandex's internal infrastructure. An agent, deeply knowledgeable in the domain, is implemented for each component. However, the tasks we solve daily can intersect domains. Therefore, we need to develop the system of agent interaction to solve the user's task with maximum quality.

Development of the metrics system and creating measurement loops Metrics are everything! As the assistant evolves, it is necessary to implement new metrics that assess new aspects of quality. It is also necessary to modify metrics to account for new user scenarios and refine old ones.

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

We expect you to

• Have worked with ML models (NLP, IR)
• Know how to plan experiments and measure their success
• Be able to write production-ready code in Python
• Are familiar with MapReduce; processing terabytes of logs doesn't scare you

Will be a plus

• Have designed and developed agent systems