Senior AI Platform Engineer

Data Science UA is a service company with deep expertise in AI and Data Science. Our story began in 2016 with the first Data Science UA Conference in Kyiv, and since then we've built one of the largest AI communities in Europe.

About the role:
We're looking for a Senior AI Platform Engineer to build the central AI platform behind a major enterprise AI transformation programme for one of our consulting clients.
A proof of concept already exists - it proves the basic plumbing works at small scale, but production-grade retrieval, evaluation, observability, and identity propagation are still ahead of us. That's where you come in: you'll take this from prototype to something real users can actually rely on.
You'll work closely with the AI Architect, who owns the overall architecture and the client relationship. Within that architecture, you'll have real ownership over how components are designed, sequenced, and built - and if something won't hold up in production, we want to hear about it, backed by evidence. This is a hands-on building role, not a meeting-heavy one.

Responsibilities:
- Build the AI gateway integration so model choice stays swappable, with no provider hardwired into the application code;
- Build hybrid retrieval that combines lexical and vector search over the enterprise corpus, including typed metadata extraction at ingestion and a query-understanding step that turns natural language into filter predicates;
- Build an evaluation harness that's genuinely maintainable — and keep it maintained — feeding eval results directly into release decisions;
- Stand up the observability layer: traces, cost, and latency per request;
- Build the governance admin panel: policy authoring, role model, audit trail, approval workflow;
- Design and implement end-user identity propagation through agent runtimes, including reproducing source-system access rules at retrieval time;
- Set the engineering standards the wider delivery team builds against, and review their work against them.

Requirements:
Professional experience
- Minimum 5 years of production Python experience, including schema design, database migrations, background job processing, testing under concurrency, and API design that remains stable beyond its first consumer;
- Demonstrated experience taking a retrieval or agent system from prototype to production, with real users and real permission models — shipped systems required, not a portfolio of prototypes;
Technical expertise
- Hands-on experience with agent orchestration, including tool calling, control flow, retry logic, and cost/latency budgeting, with the ability to describe this experience without reference to a specific framework;
- Solid understanding of evaluation methodology for LLM systems, including metric selection, dataset construction, and how results are used to gate releases;
- Working knowledge of identity and access propagation across service boundaries, and sound judgement regarding the security risks of using a service account to act on a user's behalf;
- Strong command of PostgreSQL, with the ability to treat SQL as a primary retrieval mechanism alongside embeddings, rather than a fallback;
- Experience applying prompt engineering as an engineering discipline: version-controlled, tested, and measured.

Other requirements:
- English proficiency sufficient for effective collaboration within a distributed senior team;
- Familiarity with LiteLLM as the AI gateway and Langfuse for observability.

Our current stack:
- Claude Agent SDK and MCP at the agent layer
- Postgres with pgvector
- Python throughout

We care that you've solved these problems, not that you used these specific tools. MCP experience is the strongest signal here — if you've written MCP servers rather than only consumed them, tell us about it.

Nice to have:
- Knowledge graphs and large-scale document ingestion;
- Prior experience as the first engineer on a platform that others later built on;
- Experience replacing a prototype that a team was attached to.