Senior Software Engineer AI Search & RAG

Our working language is English. We are a German-Indian team, so English is what we write, review and meet in. German is welcome but not required, also not in customer projects. Apply in English or German, whichever suits you.

What you will work on

• The architecture: you decide how our retrieval stack is built, which trade-offs we accept and which technologies the rest of the team builds on
• The retrieval core: hybrid search with BM25 and embeddings, chunking strategies, re-ranking, and the evaluation that tells us whether a change actually helped
• Agents that do real work: tool calling, orchestration, guardrails and retries, with LangChain where it earns its place and plain code where it does not
• Enterprise sources: Microsoft Graph API, SharePoint and OneDrive via OAuth 2.0, with permission-aware retrieval, so nobody sees a document they could not open themselves
• Services end to end: Python and FastAPI, PostgreSQL, Redis and Celery, Qdrant as the vector store. You design it, you ship it, you keep it running
• Operations you can trust: Docker, Kubernetes and Helm, infrastructure as code with Terraform, OpenTelemetry for traces, latency and cost per answer
• Quality you can prove: evaluation sets and regression tests for retrieval and answer quality, so releases rest on numbers instead of impressions
• The last mile: when a feature needs an interface, you build it in React rather than handing it on

• Four years or more of building software professionally, at least two of them with LLM, RAG or search systems in production, not only in prototypes
• Strong Python and solid API design with FastAPI or something comparable, plus PostgreSQL and Redis
• Retrieval fundamentals: embeddings, a vector database such as Qdrant, hybrid search with BM25, re-ranking, and an opinion on how to measure all of it
• Agents in practice: tool calling, function schemas, evaluation, and a healthy distrust of anything that only works in a notebook
• Containers in production: Docker and Kubernetes; Helm and Terraform are a plus, not a hurdle
• Enterprise integration: OAuth 2.0 and APIs like Microsoft Graph, with an eye on data protection in an EU context

Nice to have, not required: Celery at scale, OpenTelemetry, React, on-premise or EU cloud deployments, experience with the EU AI Act, open-source work we can look at.

• 80,000 to 95,000 € per year, depending on what you bring, stated openly from the start
• Short paths: you work directly with the founders, decisions happen the same day
• Every model, plus our own platform: you work with OneMachine and the leading language models and may use them for your own projects
• Real users from day one: paying customers, real documents, immediate feedback
• A start-up you can shape: you are early enough for your decisions to stay in the product
• Hybrid work at the K67 Tech Hub, Kasernenstraße 67 in Düsseldorf, a few steps from Königsallee

We are MindsMachines, an AI company in Düsseldorf. We build OneMachine, our own data-sovereign AI platform: an enterprise workspace with access to all leading language models, automated workflows on an open-source base and retrieval over a company’s own documents. It runs inside our customers’ tenants, which means permissions, traceability and latency are not optional extras. Our customers are mid-sized manufacturers, logistics companies and food producers. We have paying customers and partners from universities and industry associations.

You would be the person who makes search and agents actually work on messy enterprise data: SharePoint folders nobody cleaned up, contracts, specifications, ERP exports. Not a demo, but something people use every day.