We are building a platform for developing AI agents and developing one of the most useful internal tools — DeepAgent. It is comparable in task class to OpenAI, Perplexity, and Google, but specializes in working with Yandex's internal knowledge and infrastructure.

DeepAgent helps employees find answers to questions that cannot be solved using external search engines: understanding large codebases, documentation, tickets, logs, services, and internal processes. We are developing the Deep Research system: teaching the agent to search for information in all available sources, read and interpret data, combine it, build action plans, and improvise when performing tasks. At the same time, we are fine-tuning our own and open-source LLMs, which are optimized for agent scenarios and working with internal data.

The area is rapidly developing: a significant part of the tasks to be worked on has appeared very recently.

What tasks await you

Agent Development You will train the agent to solve complex engineering tasks: help developers debug code, understand project architecture, analyze incidents, navigate internal infrastructure and services. You will also design the agent's behavior, formulate instructions, work with data, and evaluate answer quality.

Use Case Scenario Development You will prepare and curate data for agent training, formulate and conduct experiments, fine-tune and configure LLMs for agent scenarios. You will develop new agent skills, build training, inference, and quality assessment pipelines. You will also engage in reinforcement learning to improve the agent's robustness, usefulness, and work quality, and monitor its behavior in production, iteratively improving it.

We expect you to

• Write confidently in Python
• Actively use Deep Research and code agents: Cursor, Cline, Roo Code, or equivalents
• Have fine-tuned your own and open-source LLMs or have a good understanding of approaches to fine-tuning and reinforcement learning in agent scenarios
• Work with data and experiments, quickly test hypotheses and improve model quality
• Quickly immerse yourself in new domains and complex systems
• Are ready to actively develop in the field of AI agents and research systems