Senior NLP/LLM Engineer

Sber seeks a Senior NLP/LLM Engineer in Moscow (office/hybrid, full-time). Develop an AI-copilot for premium client managers; design end-to-end LLM pipelines, RAG, and multi-agent systems.

AI management team (B2C client experience development block) is looking for an NLP/LLM engineer to develop an AI-copilot for premium client managers. The team is moving from classic NLP solutions to adaptive multi-agent systems. Responsibilities: - Design and implement end-to-end LLM pipelines, including advanced RAG systems and autonomous AI agents (planning, function calling, multi-agent orchestration). - Fine-tune and adapt models: SFT/RLHF/DPO, LoRA/QLoRA for business tasks; decide when fine-tuning is needed vs prompt engineering or RAG. - Maintain quality culture: implement product and ML metrics (RAGAS, DeepEval, LLM-as-a-Judge), automate regression testing. - Write clean, documented, testable Python code and design API contracts for ML components. - Participate in code reviews, shape team technical standards, work directly with product and stakeholders.

Experience in DL/NLP from 4 years, including training and fine-tuning models on PyTorch, with at least one year of practical LLM experience. - Practical experience building AI agents and LLM pipelines using LangChain/LangGraph. - Deep understanding of modern LLM architecture: attention, MoE, RoPE, KV-cache, alignment methods (RLHF/DPO), principles of efficient inference. - Strong engineering culture: Python (asyncio, pydantic, FastAPI), clean testable code (pytest), asynchronous microservices; understanding of Docker/K8s, CI/CD for ML. - Evaluation and monitoring: confident with LLMOps tools (Arize, LangSmith, W&B, Phoenix), experience creating custom eval pipelines. - Data work: experience preprocessing complex data for datasets, knowledge of SQL.

Full-time employment Work format: office or hybrid (Moscow) Company: Sber