Python Developer of LLM Services and AI Agents (Junior)

Metran Industrial Group seeks a Junior Python Developer for LLM services and AI agents. Hybrid or office in Chelyabinsk, full-time. 1-3 years experience.

Develop LLM services and AI agents in Python (FastAPI). Build RAG pipeline – knowledge base (regulations, accounting policies), vector search, answers strictly from sources. Connect LLM to internal services – 1C (exports, HTTP services), databases, n8n, AnythingLLM. Bring services to production – API, error handling, security, observability (monitor costs and responses). Support and improve agents – update after model or prompt changes, fix incidents, rollback versions. Follow LLM/agent trends and propose applicable improvements.

Python – confident knowledge, typing, async/await, OOP, unit tests, Git. Commercial experience from 1 year or strong pet projects – assessed by code, not resume line. Experience calling LLM via API (OpenAI-compatible) – prompting, structured output, tool calling. Pet projects count. Understanding LLM limitations – hallucinations, non-determinism, cost, latency. Honest 'I don't know' is better than fabrication. Understanding agent loop 'model → tool call → result' – and when an agent is not needed, a simple pipeline suffices. Basic FastAPI (or similar framework) and SQL. Nice to have: RAG specifics – embeddings, vector stores (pgvector/Qdrant/FAISS), chunking, faithfulness. Observability – query tracing (steps, tools, costs). Langfuse/Phoenix/OpenTelemetry. Evals – test cases and regression runs on prompt/model changes. MCP (Model Context Protocol) – our agents connect to 1C via MCP. Security of financial data – prompt injection, read-only by default, secrets in env, limits, logging. Docker, CI/CD, Linux at working level. Local/self-hosted models (Ollama, vLLM) – our main platform is Cloud.ru. n8n, AnythingLLM. Knowledge of 1C, finance/accounting – plus.

Full-time employment. Schedule: 5/2. Work format: hybrid or office in Chelyabinsk.