Description
About the Direction
We develop LLM and agentic solutions for trade finance and factoring products: from rapid MVPs and pilots to industrial implementation. The focus is on agentization and autonomous processes within product scenarios.
Projects
- Comprehensive solutions for clients: complex, hyper-personalized offerings that provide a complete solution to a company's (client's) complex task, including an optimal plan to achieve the desired outcome and the necessary products and services for implementing such a plan.
- LLM/Agentic solutions in trade finance: designing and implementing AI agents (working with documentary operations), gradually transitioning to greater autonomy with controlled risk.
- Related NLP tasks (within agent development): classification / entity extraction / summarization / knowledge base search, etc.
Technologies and Tools
- Python, PyTorch, Transformers
- OpenShift, Docker, Kafka
- Spark, Hadoop
- Agentic/RAG stack
- LLMOps/MLOps practices (observability, testing, releases).
Responsibilities
LLM-focused solutions:
- designing and implementing strategies for adapting LLMs (prompting, fine-tuning, LoRA, RLHF) to product specifics
- designing and developing data processing pipelines (RAG, agentic systems, semantic search)
- designing skills and training LLM and NLP/Classic ML models to implement business tasks
- optimizing LLM performance in production (latency, cost, accuracy).
Production Engineering:
- deploying DS models to production using MLOps practices (CI/CD, monitoring, A/B testing)
- integrating solutions with external APIs, working with vector databases, search engines
- designing fault-tolerant systems for processing confidential data
- working with SQL/NoSQL DBs.
Leadership and Expertise:
- participating in setting technical requirements and interacting with business clients
- collaborating with product experts, system analysts, and the client side to elaborate requirements and solution options for tasks
- analyzing risks and finding compromises between model quality, speed, and cost.
Requirements
- experience: 3+ years in DS/NLP, including 1+ year working with LLMs, production experience.
- experience building RAG systems, agentic pipelines, and services based on LLMs
- deep expertise in adapting LLMs: SFT, RLHF, LoRA, prompt engineering
- experience working with LangChain/LangGraph libraries
- willingness to both write code, build pipelines, train models, and write documentation, design systems, and prepare specifications for models, data, pipelines
- confident work with infrastructure: Docker, Kubernetes, cloud platforms
- understanding of MLOps: CI/CD, data drift monitoring, logging
- experience transforming business tasks into technical requirements.
Will be a plus:
- experience building autonomous processes (decisioning/workflow) with risk control
- knowledge of LLM safety/security (prompt injection, data exfiltration, guardrails)
- experience optimizing costs and latency: model selection for scenarios, caching, batching.
Conditions
- comfortable office at 19 Vavilov St., near Leninsky Prospekt metro station
- work format - hybrid
- opportunity to participate in internal and external IT conferences, a wide selection of training courses at the Corporate University
- annual salary review and annual bonus
- extended VHI, preferential insurance for family, and corporate pension program
- free SberPrime+ subscription, discounts on products from partner companies
- referral bonus for recommending friends to the Sber team.
