Senior Backend Engineer (AI Platform)

As a Backend Engineer you will build the services our products and client solutions run on. The center of gravity is LLM infrastructure: the layer that puts one governed front door over many models, hosted and self-hosted, and answers the questions enterprises actually ask — who used which model, for what, at what cost, under whose rules. The platform's first customers are our own client-facing engineers, the people delivering client outcomes with what you build, so requirements arrive concrete, feedback arrives same-day, and the people you support are in the same meeting.

- Own the LLM gateway: routing, metering, budgets, guardrail composition, and provider/backend adapters across hosted and self-hosted models.

- Manage the retrieval and knowledge service: connectors, chunking, hybrid search and reranking, grounded answers, and the MCP surface.

- Build the eval harness that turns retrieval tuning from judgment into evidence, with confidence intervals and per-segment breakdowns.

- Own the async framework — its public API, rate-limit algorithms, and admin tooling — and lead its evolution into durable execution for agent runs.

- Design authorization across the whole surface: token hierarchies, capability grammars, and per-tenant isolation tested as a regression, not asserted in a doc.

- Own your own definition of done on everything you ship: evals, traces, cost accounting, and the dashboard that would wake you up.

- 5+ years of deep, production async Python — cancellation scopes, streaming lifecycle, and connection pooling

- Postgres as infrastructure — comfortable reasoning about MVCC, advisory locks, and vacuum discipline, with Redis/Dragonfly used (and not used) where it belongs; real distributed-systems experience, not just familiarity.

- Experience shipping API surfaces other engineers build on — versioning, idempotency, error vocabularies, migration discipline, and documentation to match.

- Depth in at least one of LLM gateway concern (routing, metering, guardrails, provider failover) or retrieval engineering (hybrid search, reranking, eval methodology, audit-ready RAG), with credibility on the other.