Description
We are a research unit within Sber responsible for the "human" aspect: behavior, psychology, trends, changes, and features of perception and thinking.
Our key task is to research and communicate what is happening and will happen with people, and how to translate these changes into products, communications, and HR processes to make them better.
Working at the Laboratory is an opportunity to engage with almost any business unit of the Sber ecosystem, Sber's big data, and a client base of over 100 million people.
We are looking for an "AI Generalist". A person who will quickly (1-2 weeks) vibe-code prototypes for various projects based on the Laboratory's hypotheses and business requests.
Our technology stack:
- ML/DL: HuggingFace Transformers, SentenceTransformers, CatBoost, XGBoost
- LLM & Vector Search: Local models (Llama 3, Mistral), API (OpenAI), FAISS
- Data & Infra: ClickHouse, Docker, FastAPI, Postgres, RabbitMQ
- MLOps & Orchestration: MLflow, W&B, Prefect, Dagster, Ray, Prometheus, OpenTelemetry
- Tools: Hydra, Poetry.
Responsibilities
- Developing solutions on how to apply LLMs for data analysis, accelerating research, and creating new tools (e.g., parsing data, generating hypotheses, summarizing various research studies and processing them for target tasks)
- Vibe-coding prototypes for various projects based on the Laboratory's hypotheses and business requests
- Developing simple AI agents, RAG knowledge bases, graph databases
- Essentially, you will be an AI and vibe-coding evangelist who assembles various solutions and consults colleagues
Requirements
AI-native development and tools:
- Practical experience with AI-oriented IDEs and agents: Cursor, GitHub Copilot, Claude Code, Replit Agent
- Designing architecture before code generation
- Experience with large-context projects
- Experience in quickly localizing and fixing LLM logic errors
- Experience implementing guardrails and fail-safe mechanisms
- Ability to make LLMs consistently return valid JSON
- Advanced work with prompts (few-shot, CoT, ReAct, etc.)
- Experience with N8N and Make
LLM and Retrieval Architectures:
- Experience building production solutions using LLM APIs (OpenAI, Anthropic, etc.)
- Working with embedding models and vector databases
- Understanding the differences between vector search and BM25
- Experience applying various chunking strategies
- Experience with LangChain / LangGraph
- Understanding the impact of hyperparameters (Temperature, Top-P, Top-K, Frequency/Presence Penalty) on determinism and hallucinations
Will be a plus:
- Controlling the cost and latency of AI calls
- Ability to quickly bring "vibe-code" to production quality
- Ability to design and build REST and GraphQL APIs
- Production deployment and operation of applications in modern cloud environments
- Understanding of CI/CD, monitoring, logging, and observability
- Deep interest in the topics of AI, agents, vibe-coding
- Technical education (programming / engineering)
- Ability and, most importantly, desire to quickly deliver prototypes without extending development timelines
- Willingness to delve into the meanings of tasks, products, and research – often the input will be an abstract requirement specification that needs to be refined into a working solution
- Developed soft skills in communication (will need to translate from technical language to business language)
- Tracking and analyzing cutting-edge advances in AI and LLMs, promptly adapting innovations to business tasks.
Conditions
- Comfortable modern office in Moscow, near Kutuzovskaya metro station
- Opportunity to choose a convenient schedule – office/hybrid
- Annual salary review and annual bonus
- Corporate gym and relaxation areas
- More than 400 educational programs from SberUniversity for professional and career development
- Private health insurance, preferential insurance for family, and corporate pension program
- Flexible mortgage discount equal to 1/3 of the Central Bank's key rate
- Free SberPrime+ subscription, discounts on products from partner companies.
