Description
Team Description:
Our team develops comprehensive AI solutions (Classic NLP, LLM, AI-agents) for the key products and processes of the "Strategy and Development" Block. We are always at the forefront of technology development and try new things – we were the first at Sber to develop a prototype of a multi-agent system for handling inquiries, which made agent-based solutions one of the most in-demand areas in the bank.
Our technological focus goes beyond AI agents: we solve tasks of classification, clustering, matching, domain adaptation (Metric Learning, PEFT), using SFT when necessary. The outcome of our work is not a separate e2e pipeline, but production-ready multiservice architectures integrated into Sber's internal surfaces.
Recent projects: a multi-agent system for analyzing organizational effectiveness (presentation at AI Journey 2025), a multi-agent pipeline for analyzing documents on organizational changes, an AI-agent-Copilot for setting and monitoring goals. All directions are actively growing and receiving direct support from the bank's management.
Main areas of activity:
- We formulate recommendations for improving the efficiency of teams, products, and departments based on classification, clustering, and topic modeling using digital footprints (Jira, meetings, emails, etc.)
- We conduct comprehensive performance analysis within the framework of scenario modeling of activities to achieve the bank management's goals
- We implement pipelines for processing internal documents of arbitrary length to build recommendations for working with them and accelerating organizational changes
- We identify global trends and analyze their impact on the number of bank roles for Sber's Strategy
- We analyze organizational goal graphs (connectivity, cascading) and also recommend ambitious goals considering the context and priorities of the Strategy
- We expand the domain adaptation direction to enhance streams of semantic search, ranking, and other NLP downstream tasks
- We participate in the development of the global AI agents direction and regularly use modern LLM-based approaches in our work (External Tools, Reasoning, Reflection)
- We test hypotheses of any complexity to obtain Data-driven insights for preparing strategic sessions for the bank's management
Our global priorities:
- Development and implementation of AI solutions (Classic NLP, LLM applications, AI agents) to improve the efficiency of the bank's priority strategic processes with potential for use in the external market
- Creation of SotA solutions considering the specifics of the bank
Why choose us:
- Opportunity to use advanced AI technologies and the bank's platforms
- Participation in the development of innovative services for the strategic block, which bring real benefit to the processes and products of the entire bank and quickly come to the attention of key managers
- Opportunity to participate in international projects and conferences on AI and ML
- Work in a friendly team of professionals focused on achieving the most ambitious goals and constant development
Responsibilities
- Development and implementation of AI services (Classic NLP, LLM applications, AI agents, dialogue systems) from the MVP stage to PROD (CRISP-DM)
- Solving NLP tasks: Preprocessing, Classification, Summarization (Ext/Abst), Sentence Compression, NER, Semantic Search, Clustering, etc.
- Creating multi-agent pipelines based on frameworks for working with LLM (LangGraph, LangChain/GigaChain, LlamaIndex, etc.)
- Adapting and training language models based on internal and external data (Prompt Tuning, RAG, PEFT, SFT)
- Indexing and ranking of text documents
- Interaction with the business customer to identify requirements and independent task formulation
- Optimization of AI services in a production environment
- Participation in model validation and automonitoring, conducting A/B testing
Requirements
- Higher technical education in the field of computer science, applied mathematics and informatics. The most preferred: HSE, MIPT, MSU, Skoltech
- Experience in developing NLP models and recommendation systems
- Understanding of the model lifecycle (CRISP-DM)
- Ability to translate business problem statements into ML problem statements, competent interpretation of the obtained results
- High proficiency in Python core and SQL
- Knowledge of machine learning frameworks, libraries, algorithms: XGBoost, CatBoost, PyTorch, TensorFlow, Transformers
- WEB frameworks: FastAPI (async methods), Flask, etc.
- Knowledge of neural network architectures: RNN, LSTM, transformers (BERT, BART, T5)
- Knowledge of agent architectures (ReAct, Blackboard, Multi-agent, etc.)
- Knowledge of frameworks for working with LLM (LangChain/GigaChain, LangServe/GigaServe, LlamaIndex, etc.)
- Containerization: Docker, OpenShift
Conditions
- Flexible hybrid (discussed individually)
- Modern IT office near Moscow-City with a fitness center
- Mortgage with benefits for the employee and preferential lending terms
- Free SberPrime+ subscription
- Discounts on products of partner companies
- Health insurance from day one and preferential insurance for family members
- Corporate pension program
- Company-paid training: online courses at Sber's online school and unlimited access to the library, training at the Corporate University, trainings, meetups, and the opportunity to obtain a new qualification
- Largest DS&AI Community — over 600 DS in the bank, regular exchange of knowledge, experience and best practices, interactive lectures and master classes from leading universities and experts of technology companies, digest of the latest developments in the field of DS&AI and reports from the world's largest conferences