AI engineer

Мы в поиске Senior AI инженера в международную компанию. О вакансии Компания создает слой искусственного интеллекта на своей платформе: встроенный в приложение помощник, который отвечает на вопросы и выполняет действия, аналитические помощники на базе нашего хранилища данных, сервер MCP, предоставляющий более 100 платформенных инструментов для помощников с искусственным интеллектом, и мультиагентные системы, автоматизирующие диспетчеризацию. В качестве старшего инженера по искусственному интеллекту вы будете проектировать, внедрять и эксплуатировать эти системы в производственной среде. Стек — Python на Google Cloud (Vertex AI, Cloud Run, BigQuery) с несколькими поставщиками моделей (Gemini, Claude).

• Build and own the in-app AI copilot that answers questions and performs actions on routes, orders, drivers and vehicles;
• Extend our MCP server (100+ tools) so AI assistants can work with the platform safely;
• Design multi-agent systems for automated dispatching, with agent routing and human approval for risky actions;
• Build AI analytics: natural-language questions and insight reports over our BigQuery data;
• Own the AI platform layer: prompt and model versioning, model routing, caching, guardrails, tracing, cost and latency;
• Write design docs and ADRs; work with the PO, backend (PHP/Python) and frontend teams; share AI practices across the team;

• 5+ years of commercial software engineering, Python as the primary language (async, typing, testing; FastAPI or similar);
• 2+ years building and shipping LLM-powered products to production for real users;
• MCP servers: design and build MCP servers and tools on top of existing REST APIs (auth, multi-tenancy, rate limits, error handling);
• AI chats: production conversational assistants with multi-turn context, streaming, tool calling and structured outputs;
• Agents: tool-using agents with planning, state, retries, termination conditions and human-in-the-loop for risky actions;
• Agent systems: multi-agent orchestration (supervisor/worker, handoffs, shared state) with LangGraph, Google ADK, OpenAI Agents SDK or custom code;
• Agent routing: intent classification and routing between agents, tools and models by task, cost and latency, with fallbacks;
• RAG: chunking, embeddings, vector and hybrid search, reranking, grounded answers with citations;
• System prompting and context engineering: versioned system prompts, tool descriptions, few-shot examples, context window management, prompt-injection defense;
• Modern AI architecture: model abstraction across providers (Gemini, Claude, OpenAI), caching, guardrails, tracing and observability, cost and latency control;
• Cloud and delivery: GCP or AWS, Docker, CI/CD (GitHub Actions), monitoring;
• Security for AI actions: tenant isolation, least-privilege tools, audit logs;
• AI-assisted development as a daily habit (Claude Code, Cursor, Copilot);
• English B2+ (written and spoken); clear design docs and ADRs;
• Будет плюсом: Production evals: golden datasets, CI regression gates, LLM-as-judge, A/B tests of prompts and models; Google Cloud: Vertex AI (Gemini, Agent Engine, prompt management, context caching), Cloud Run, BigQuery, Pub/Sub; Text-to-SQL and analytics agents over BigQuery-scale data; Workflow engines: Google Workflows, Temporal, or event-driven pipelines (Pub/Sub, Kafka); Long-term (semantic) memory for agents, scoped per tenant; Frontend for AI UX: TypeScript with Vue or React/Next.js (streaming chat UIs); Kubernetes (GKE), Helm, Terraform; Ability to read PHP/Laravel code (our core platform APIs); Fine-tuning or serving open-weights models (Gemma, Llama, Qwen; vLLM); Optimization tools: OR-Tools, VRP heuristics; Published MCP servers, connectors or open-source work in AI tooling

• Semi-flexible working hours;
• Completely remote work;
• Interesting projects to work on;
• Exposure to a variety of industries and technologies.