AI engineer for financial advisory

We're hiring an AI Engineer to lead the design and build of production agents and LLM features for a specialized financial advisory firm. You'll own the work end to end, from architecture through the unglamorous production hardening that decides whether an agent actually holds up: streams that reconnect cleanly, tool calls that never fire twice, structured output that validates every time.

• Lead the design and development of agents and LLM features end to end, from architecture through production hardening
• Architect core agent-runtime capabilities: session lifecycle, SSE streaming with reconnect and de-duplication, crash-safe tool dispatch, and structured and streaming model integration
• Build out the provider-agnostic LLM client (text, JSON-schema structured output, streaming tool use) and the prompt registry behind it, with templating and versioning
• Own the evaluation frameworks and quality gates the team relies on to ship model and prompt changes with confidence
• Take on the ambiguous, high-impact problems, like streaming reliability, tool-call correctness, and cost versus latency trade-offs
• Telemetry over LLM requests and background jobs, so token usage, latency, cost, and outcome quality are always visible
• Compose tools and actions across both self-hosted and cloud-sandbox execution models
• Work on the integration layer and its mock seams, plus columnar analytics where agents reason over data
• Mentor engineers on craft, conventions, and judgment. A big part of this role is setting the patterns others follow.

• Mid-level or senior experience building production LLM or agent systems, including tool/function calling, structured output, prompts, and streaming, and a working knowledge of how each one fails
• Strong systems engineering in Python (or a comparable backend language) with sound architectural judgment
• Rigorous testing discipline: dependency injection, boundary mocks, and tests that assert behavior rather than implementation
• Experience building evaluation pipelines or quality gates for model and prompt changes
• A mentoring instinct and a track record of raising the quality of the engineers around you
• Based in the UK or Germany and comfortable working fully remote
• Будет плюсом: Hands-on Anthropic API experience: Claude tool use, the Managed Agents API, MCP, or Skills; Familiarity with actor or agent frameworks and columnar analytics tools like DuckDB; Background in finance, accounting, audit, or another regulated domain

• You'll join a small, senior engineering group inside a firm with deep finance and accounting expertise that's now turning that expertise into software. The bar is production quality, not prototypes, and the problems are the ones most teams skip: correctness, reliability, and cost at scale.
• If you want your judgment to shape how an agent platform gets built, and to see that work run against real financial workflows, this one is worth a conversation.