Python developer of AI services (LLM / RAG)

QSOFT is hiring a Middle Python developer for AI services (LLM/RAG), full-time, remote-friendly, focused on FastAPI, semantic search, and agent workflows.

Develop HTTP API and business logic of two async Python services on FastAPI. Design and implement RAG pipelines: retrieval, query rewriting, multi-hop search, context collection, structured output, source citation, and partial answer handling. Develop agent scenarios on LangChain, LangGraph, and deepagents: tools, middleware, call limits, cycle protection, and response contract control. Improve semantic search quality: chunking, dense/sparse retrieval, Qdrant filters, reranking, RRF, and token budget. Develop event-driven indexing of articles, files, and structured content via Kafka. Ensure correct multitenancy and result filtering by user, group, and community permissions. Integrate and replace LLM, embedding, and reranking providers, including OpenAI-compatible API, vLLM, and Ollama. Work with PostgreSQL, Redis, and Qdrant, design data models and Alembic migrations. Diagnose quality, performance, and reliability issues at the boundaries of external services. Develop automated tests and quality gates, maintain reproducible local and CI environments. Improve observability: structured logs, technical metrics, OpenTelemetry, and LLM/RAG tracing via Langfuse. Participate in task decomposition, code review, and architectural decisions.

Confident Python and practical backend development experience. Good understanding of async programming: event loop, concurrent I/O, timeouts, cancellation, resource management, and graceful shutdown. Experience with FastAPI or comparable Python framework. Practical experience with PostgreSQL and SQLAlchemy: transactions, concurrency, indexes, migrations, and query diagnostics. Experience integrating external services and designing resilient interactions: retry/backoff, timeout, idempotency, concurrency limits, and partial failure handling. Understanding of event-driven architecture and Kafka consumer semantics: consumer groups, offset commit, rebalance, redelivery, and idempotent processing. Understanding of RAG: chunking, embeddings, retrieval, reranking, context window, LLM limitations, and hallucination reduction. Ability to write unit and integration tests for async code, separating business logic from external infrastructure. Ability to navigate an existing codebase, find root causes, and propose minimal, testable changes.

Full-time employment. Remote-friendly format. Opportunity to work on AI services.