Description

Our team (70+ people) creates the technological foundation for integrating GenAI into the bank's ecosystem. We are developing a RAG platform, an agent execution framework, and systems for processing unstructured content (documents, communications). Our services wrap base LLMs in ready-to-use interfaces, enabling the business to quickly create their own agents and GenAI applications.

We ensure a full development cycle (E2E): ML engineers and backend developers work together on service implementation, in conjunction with frontend developers and designers. We have our own labeling team for quality assessment and dataset collection, which we use to fine-tune domain models for banking scenarios.

Scale of tasks:

Currently, the portfolio includes over 200 implemented scenarios.

Planned — implementation of 300+ more scenarios, including creating a procedural memory layer for agents, specialized search agents, and automatic GenAI quality assessment systems (Auto-Eval).

Development of Self-Service tools (UI sandboxes), where bank teams can test hypotheses before integration.

What you will be doing

We are looking for engineers who are interested not only in R&D but also in creating full-fledged product systems capable of learning from user feedback and solving specific business problems.

Responsibilities

• Development of GenAI services: Creating and improving RAG systems (including Agentic RAG), Agent Observability tools, document processing services
• Model training: training and fine-tuning multimodal models in Russian (OCR, Embeddings, Rerankers, VLM)
• MLOps and Data Management: Organizing and automating the data labeling process (from collection to quality control of crowdsourcing)
• development of autonomous pipelines for improving model quality using LLM-as-a-Judge
• Engineering: releasing new models and services into production environments.

Requirements

• Deep understanding of LLM: experience working with large language models, Prompt Engineering, Fine-tuning LLM/VLM
• Engineering culture: proficient knowledge of Python 3, knowledge of OOP, SOLID principles. Good knowledge of algorithms and data structures.
• LLM Stack: experience with LangChain, LlamaIndex, or other frameworks for developing agents and multi-agent systems (MAS)
• Vibe Coding: ability to automate one's own routine processes using AI tools
• Mindset: desire to constantly study new SOTA approaches, models, and technologies.

WILL BE A PLUS:

• Active profile on GitHub with quality code
• achievements (medals) on Kaggle
• participation in Open-Source projects related to LLM/NLP.

Conditions

• comfortable modern office near Kutuzovskaya metro station
• hybrid work format
• corporate gym and relaxation areas
• more than 400 educational programs from SberUniversity for professional and career development
• extended health insurance, preferential family insurance, and corporate pension program
• mortgage for employees is more favorable by up to 4%
• free SberPrime+ subscription, discounts on products from partner companies
• reward for referring friends to the Sber team.