AI Engineer

💡About the team & role

You’ll join the AI Competence Center of a leading Ukrainian bank — the team that designs, builds, and ships production AI systems the rest of the bank relies on. We are building production-grade AI agents for a regulated banking environment. Your core mission is code development of AI agents on modern frameworks, from prototype to governed production deployment (OnPrem and on AWS). Because we operate in a regulated financial environment, guardrails, PII handling, and model governance are part of how we build — not an afterthought.

We are looking for a strong agent engineer who is experienced, willing to take on challenges, drive projects forward, share knowledge with colleagues, and train others.

It is better to examine it firsthand  ПУМБ — YouTube

🎯 What you will do

Agentic Systems

• Design, build, and deploy stateful AI agents on modern agent frameworks (LangGraph, LangChain, CrewAI, AutoGen, OpenAI Agents SDK, Google ADK, Semantic Kernel, or similar) — we expect real depth in one or two of them and the ability to pick up others quickly.
• Implement the core agent primitives in production: multi-agent orchestration, tool/function calling, MCP, memory, routing, planning, structured outputs and human-in-the-loop flows.
• Develop RAG systems using hybrid retrieval, metadata filtering, reranking, and grounded generation over the bank’s data.
• Build automated evaluation suites, golden datasets, and regression tests.
• Implement tracing, production monitoring.
• Own the full agent lifecycle — build, deploy, monitor, and improve — including evaluation, observability, and reliability (task completion, groundedness, tool-call accuracy, latency/token-cost optimization, hallucination mitigation, drift detection and policy violations).
• Deploy and operate agents on the OnPrem AI platform (additionally as plus — AWS Bedrock/SageMaker or equivalent), packaged as containerized services.
• Build in guardrails, PII handling, and human-in-the-loop controls appropriate for a regulated bank, and make agent behavior auditable and governable.
• Share best practices within the Agentic AI chapter.
• Enable business domain teams (technical and non-technical) to build and support their own AI agents on the AI platform, so development and support can be decentralized.

Shared AI services

• Contribute to shared AI services other bank teams integrate: MCP servers, AI model REST APIs.
• Build and maintain MCP servers, secure tool adapters.
• Containerize and deploy services on-premises.
• Produce high-quality documentation, runbooks and integration examples.

🛠️ What We’re Looking For

Must have

• 1-2 years of professional experience in Agentic AI and delivering at least one production agentic system.
• 5+ years of professional experience in software engineering, ML engineering, or applied AI.
• Python — strong, production-grade including typing, testing, packaging, asynchronous programming, profiling, and maintainable architecture.
• Hands-on production experience with at least few modern agent framework (LangGraph / LangChain / CrewAI / AutoGen / OpenAI Agents SDK / Semantic Kernel / Google ADK).
• A strong understanding of agent architecture: state, memory, tools, planning, routing, retries, idempotency, and human approval.
• Practical experience with RAG, embeddings, vector or hybrid search, and retrieval evaluation.
• Experience building production APIs, microservices, or distributed backend systems.
• Experience with Docker.
• Solid practical experience and knowledge with evaluation and observability tooling for AI workflows, Agents, Multi-Agents systems, context retrieval and regression testing for probabilistic systems: DeepEval, RAGAS, Langfuse.
• Experience building MCP servers
• Strong knowledge of Transformers architecture (encoder, decoder, attention) and embeddings
• Awareness of guardrails, PII handling, and responsible-AI practices in a regulated setting (you don’t need to be a model-risk specialist — you need to build with these in mind).

Nice-to-Have

• Working knowledge of TypeScript for SDKs, integration services, or full-stack components.
• Experience in model training/fine-tuning at least one of the following areas: Computer Vision, NLP, ASR, TTS, or classical ML
• Experience of model versioning, registry, monitoring, drift detection, and retraining workflows
• Experience with AWS
• MLflow, Weights & Biases, or an equivalent model-lifecycle stack
• PyTorch, TensorFlow, Hugging Face Transformers, or scikit-learn.
• Experience in banking, fintech, or another regulated industry.
• Knowledge of model risk management, Responsible AI, or adversarial AI testing.

Tech stack

• Languages: Python (primary), TypeScript (nice-to-have)
• Agent frameworks: LangGraph, LangChain, OpenAI Agents SDK , CrewAI, AutoGen, Semantic Kernel, ADK
• Search and data: Qdrant, Milvus
• Serving & APIs: FastAPI, REST, gRPC, MCP servers
• Messaging: Kafka or RabbitMQ;
• Evaluation & observability: Langfuse, OpenTelemetry, Prometheus/Grafana, RAGAS, DeepEval
• Infrastructure: Docker

Expected outcomes

• Deliver production agentic use cases with automated evaluation, guardrails, and monitoring.
• Build a reusable AI or MCP service.
• Establish a baseline agent-evaluation and regression pipeline.
• Define patterns for human approval.
• Produce documentation.
• Reduce the time required to move AI use cases from discovery to production through reusable platform capabilities.
• Share experience to chapter and business domain teams.

🌟 Why 7,000 Employees Have Chosen Us

• Growth every day: structured career development, internal training programs, and opportunities to gain new skills.
• A supportive, inspiring team: a professional environment where your ideas matter and collaboration is the norm.
• Flexibility that works for you: hybrid format, flexible schedule, and the ability to balance productivity with comfort.
• Wellbeing matters: table tennis, psychological support, and legal assistance to help you stay balanced.
• Meaningful impact: participation in social initiatives that make a real difference and add purpose to your work.
• A culture of trust: mistakes are treated as learning opportunities, and achievements are celebrated together.
• Modern tools & technologies: everything you need to work efficiently and enjoy the process.
• The vacancy is open to defenders of Ukraine

Відгукнутись на вакансію