Description
We are the R&D team of Sber AI, creating the next generation of agent systems. Our developments form the foundation of AI assistants, coding agents, and analytical solutions. We don't just integrate LLMs—we design architectures where agents interact, reason, and solve business tasks autonomously. We work at industrial scale, yet maintain a culture of rapid experimentation: from hypothesis to a working MVP within days.
Responsibilities
You will design and implement the core of agent systems—the backend infrastructure that turns language models into autonomous problem solvers:
- develop MCP servers to connect business systems as agent tools
- build RAG pipelines with custom logic for extraction, ranking, and context synthesis
- implement agentic capabilities in a reactive loop: memory (short- and long-term), planning, communication via the A2A protocol
- design plugins and extensions for AI agents using clean architecture (abstraction layers between the agent, tools, and business logic)
- write specifications for agent behavior (agents.md), prompts with control logic, and customize AI SWE agents
- conduct systematic analysis of business processes and design information models for agent integrations
- document the architecture of agent systems, tool contracts, and interaction scenarios.
Requirements
Basic—without this, it's impossible to work in our paradigm:
- commercial Python development experience of 3+ years with deep understanding of OOP, design patterns, and SOLID principles
- practical experience in system design: ability to define domain boundaries (DDD), design layers of responsibility, write testable code (TDD)
- confident work with relational and NoSQL databases, understanding their application in the context of agent states and long-term memory
- experience with vector databases and search technologies (implementation or integration)
- understanding of algorithmic complexity and the ability to select appropriate data structures for the task
- practical experience with OpenAI API or equivalents: not just making calls, but designing reliable pipelines with error handling, rate limiting, caching.
Will be an advantage
- experience with agent interaction protocols (MCP, A2A) or their equivalents
- experience with the huggingface/transformers stack, BERT/GPT/LLAMA models, the sentence-transformers library
- experience with text and document processing tasks, projects in the NLP domain
- experience with the LLM/AI agent stack, RAG, langchain/langgraph
- involvement in designing systems with uncertainty (handling incomplete data, decision rollbacks)
- mentoring or coaching experience
Conditions
- stable salary and social support for employees
- extended corporate health insurance from the first day of work for employees and discounted medical insurance for close relatives
- free SberPrime+ subscription, discounts on products from partner companies
- corporate pension program
- corporate training at company expense
- employee referral program: you can invite familiar professionals to the team and receive a reward of up to 100,000 rubles
- powerful hardware, additional monitors, and everything needed for productive work
- work in Agile with the best in the IT industry: 2000 product teams and the possibility of internal transfers.
