Senior AI Engineer in Bremen

• Most AI teams talk about evaluation. At Synera, you will own it: end-to‑end, in production, for multiple real agentic products used by companies like BMW, Airbus, and NASA




• You’ll join the Agentic Ants team as their second native AI engineer. You’ll build and extend agentic systems alongside the rest of the team, including new tools, sub‑agents and prompt iterations. But most importantly, you will contribute to an evaluation framework that gates every change: Golden datasets, LLM-as-judge pipelines, regression suites, production‑trace mining. Your job is to prove that the agents the team is shipping can be trusted




• You’ll not only improve our existing software but also shape the next AI‑native generation of Synera. Partnering with Danny and the wider R&D department, you’ll help the teams go deeper on customer insights data and come up with new ways to automate engineering work




• What a week at Synera could look like:




• Monday: Join our company-wide all‑hands meeting to stay in the loop, then review Langfuse traces from the weekend and flag any new failure modes worth triaging




• Tuesday: Work on the golden dataset for the supervisor routing surface, curating examples, versioning the set, and writing evaluators with Ahmed




• Wednesday: Join a cross‑team sync with QA and product to align on new eval coverage for an upcoming agent feature, then push a CI integration so eval regressions block the next PR




• Thursday: Show off your latest work to other departments in the sprint review, use your lunch break to go for a quick run, then kick off sprint planning with the Agentic Ants




• Friday: Pair with the AI and software engineers on extending our agentic system, then write the eval that gates the change before it ships




• In 6 months:




• The evaluation framework is live and in use across multiple agent, CI gates are blocking on regressions, and the team can actually trust the results




• You’ve shipped multiple meaningful changes to the agent graphs (new tools, sub‑agents, or routing improvements) that are measurably better in production




• In 12 months:




• The agents improve themselves: Production traces feed continuously back into datasets, and the improvement loop is running without manual intervention




• You co‑own at least one agent surface with the team and have become the go‑to person for evaluation methodology at Synera. The combination has measurably improved product quality

Benefits

• Powerful equipment - Synera empowers its team members with the equipment they need to do their best work.




• Company offsites - Synera is a remote/hybrid company - which is why it enjoys regrouping for two offsites per year.




• Company fitness & Jobrad - Healthy body, healthy mind. Synera enables its team members to keep up with their health!




• Development budget - Never stop learning: Synera wants its team members to grow and continuously learn new things!




• Team spirit - Synera enjoys spending time with each other and places a high value on functioning as a team - no siloes, no politics, and lots of laughs.




• Flexibility - As a remote/hybrid company, Synera offers all the flexibility of a modern company to attract exceptional talent.

You communicate clearly and push back when you disagree

You write production-grade Python (using FastAPI and PostgreSQL)

You handle reliability when it shows up in your work, implementing retries, error handling, or graceful degradation

You’ve worked with at least two of the following AI platforms: Anthropic, OpenAI, Azure Froundry, Bedrock, or Vertex AI

You resonate with Synera’s Core Values - they’re central to how we work, and we’ll explore them together in your first interview

You’re familiar with Langfuse, LangSmith, or similar tracing tools

You’ve designed and shipped agent graphs in production with LangGraph or equivalent, including supervisor / sub-agent patterns, tool design, prompt iteration

You’ve built or meaningfully contributed to an LLM evaluation pipeline: LLM-as-judge design, calibration against human labels, dataset versioning, and the statistical reasoning behind it (CIs, sample sizes, false positives)

#J-18808-Ljbffr