We build the automated evaluation and testing layer that lets companies put AI agents into production in places where being wrong is expensive.This is the engineer we put in front of the customer. Your first engagement is a German insurance venture building an AI-driven B2C insurance value chain, quote, consult and purchase all AI-driven, going live in September, alongside a second track automating their internal insurance operations.We work Claude Code first. If you do not build that way today, this is not the right role for you.What you'll doTake a vague business need, turn it into a system that works, and keep it working.Pick up signal fast. Hear one sentence in the middle of a meeting, work out what the business actually wants without being given much detail, and go deliver it.Own outcomes end to end. We drop you into a project, say "automate this manual customer workflow", and three months later it works, without anyone directing every detail.Own the customer relationship. We build the most robust software we can, but if the customer does not see the value, that software gets thrown out.Push the business for clarity where the right AI behaviour cannot be intuited, or where the data to decide it does not exist yet.What we're looking forSeveral years building and running production software that real customers depend on: APIs, backend services, data pipelines, and the operational reality of keeping them up.A track record of owning a customer-facing build from zero to production, ideally more than once.Claude Code or Codex as your daily driver. This one is a hard requirement.Around a year of hands-on work building LLM features in production. (nobody has five; claiming it is a red flag)Strong discipline around automated testing, CI/CD and monitoring. Evaluation and testing are our product, so this matters more here than in most places.Real time spent in a customer-facing role: consulting, forward deployed engineering, founding your own company, or an in-house role where you sat with the business.Pragmatic and solutions-first. No "not in my ticket" reflexes.German is a strong advantage, not a requirement. A significant share of the data you will work with is in German, as are many of our customers and their end users. Our working language is English.How we buildTypeScript end to end. Next.js and React on the front, Postgres on Supabase, AWS underneath, an agent framework (we use Mastra, any comparable one is fine), and Langfuse for tracing and evaluation.Claude Code is the primary build tool, not an accessory. That is the part we care about most. The specific tools matter far less than how quickly you become productive with them. If you have shipped serious software in another stack and you steer AI agents well, you are in scope.The setupBerlin based, hybrid. Mostly remote, around 10% travel to meet the customer. We are not able to consider applicants based outside Germany.Full time preferred, four days works, three is possible. Freelance, permanent or B2B, your choice.EUR 80,000 to 120,000 per year depending on experience. Start ASAP.How we hireA call with our CTO.A short practical challenge done with Claude Code or Codex. We grade how you steer the tool, not the outcome.Two steps, and we move fast.

Sourced from LinkedIn. Relocantly aggregates public job postings; apply on the original site.

Senior AI Software Engineer, Customer-Facing