Senior AI Engineer, LLM Agents, Tool Calling, Evals, Vertex AI, Agentic AI, Remote
Llamazing Digital Engineering GmbH
Ansprechpartner: Björn Froese
BerlinBrandenburg, Deutschland
Endkundenprojekt
ITRAGOpenAIGeminiVertex AILangChainKünstliche IntelligenzPythonMachine LearningNatural Language ProcessingLarge Language ModelsPrompt EngineeringGenerative AIBeschreibung
For agentic AI projects in industrial manufacturing in the DACH region, we are looking for experienced AI engineers for LLM agents (m/f/d) to join our expert pool. The work is 100% remote, and we welcome applications from anywhere in the world.
A typical project looks like this:
Starting point
A manufacturing company wants to handle production disruptions with an agentic workflow. When a machine fails, AI agents gather information from MES, production planning, maintenance and documentation, evaluate options (e.g. wait for the repair or move orders to an alternative machine) and present them to the responsible roles for approval. Only after human approval are the maintenance order and the rescheduling executed in the target systems. Decision-relevant figures come from traceable rules, not from the language model; every step can be interrupted and resumed and is fully logged. The solution runs on Google Cloud with Temporal as workflow engine. You own the agents: what they may query, how they back findings with sources and how they are protected against injected instructions.
Responsibilities
- Design agents for maintenance and production planning: roles, tools, limits for steps, runtime and tokens
- Define tool contracts with typed, structured inputs and outputs; every tool call is its own logged activity
- Implement structured model outputs with schemas, source references and labeled assumptions
- Implement and test protection against prompt injection from documents and tool results
- Build evaluation: test cases, reference runs, regression tests on model or prompt changes, a recording mode for demos
- Connect models via Vertex AI on a regional EU endpoint, prepare fallback and model changes
- Work closely with the workflow engineer: the agent loop runs inside the deterministic workflow
- Document prompts, schemas and evaluation results under version control
Must have
- At least 5 years of software development experience, including at least 2 years with LLM applications
- At least one LLM application with tool calling or function calling in production or in a customer pilot
- Hands-on experience with structured outputs, evaluation of LLM systems and protection against prompt injection
- Strong Python or TypeScript, version control and automated testing
- Fluent English
- At least 4 hours of daily overlap with Central European Time (CET)
Nice to have
- Vertex AI or Gemini, Anthropic Claude or OpenAI via APIs; experience with several providers
- Agent frameworks (e.g. LangGraph, Google ADK, OpenAI Agents SDK, Model Context Protocol)
- Experience with workflow engines such as Temporal
- Retrieval over technical documentation (RAG) with source attribution
- Experience with industrial or maintenance data
- Experience working in agile, distributed teams
- German language skills
Terms
- Location: 100% remote
- Duration: typically 4 to 6 months, extension possible
- Workload: 80 to 100 percent
- Start: depending on the project, from Q1 2027
- Contract: freelance, invoicing in EUR
Application
Please answer briefly:
- Which LLM application with tool calls did you most recently bring into production or a pilot (purpose, model, your role, number of users)?
- How do you measure the quality of an agent, and how do you prevent content from documents or tool results from acting as instructions?
- What is your hourly rate in EUR, where are you based (country and time zone), and when are you available?
In email communication please always include the project ID of this ad. Please do not call unless an appointment has been arranged.