Plattform AI Engineer

Guldberg GmbH; Frankfurt; On-site; EUR 90,000 - 120,000; Full time

Job summary

Guldberg GmbH in Germany seeks an experienced AI framework engineer to design and implement reusable components for agentic systems across the software stack.

You will author architecture records, define evaluation procedures, and train teams. This role emphasizes collaboration with Architecture, Security and Compliance to ensure robust, scalable solutions.

Qualifications

• Hands-on experience using AI tools in software development (12+ months) e.g., Kiro CLI, Copilot, Claude.
• Several years of IT consulting and knowledge transfer.
• Strong understanding of LLMs, context windows, embeddings, RAG, tool calling, agents, guardrails, prompt injection, evaluation, hallucinations, token costs.
• Ability to explain technical solutions so others can use them independently.
• Practical Scrum experience with backlog, Definition of Done, Jira.

Responsibilities

• Design and implement reusable framework components for agentic AI systems, including schemas, contracts, automated tests, and documentation.
• Create Architecture Decision Records and arc42 documentation, including rejected alternatives.
• Establish evaluation procedures for agentic workflows, including test cases, quality metrics, comparison datasets, and reassessment after model changes.
• Implement blocking quality gates in GitLab CI/CD pipelines.
• Maintain a versioned and searchable component registry and define a contribution process for other teams.
• Assess teams’ enablement needs and provide training, workshops, guided use‑case sessions, hackathons, office hours, and demonstrations.
• Train teams in the tool stack, AI tools, agentic working methods, quality assurance, limitations, and common AI failure modes.
• Create and maintain training materials, guidelines, reusable examples, and reference cases.
• Support teams until they can use the framework independently and act as a contact for user questions.
• Coordinate with Architecture, Security, Data Protection, Compliance, and Operations.
• Participate in backlog refinements and ART planning, including effort estimation and task slicing.
• Ensure compatibility across macOS, Linux, and Windows environments.
• Monitor the costs and runtimes of agentic workflows and recommend suitable alternatives.
• Define how AI-generated changes are created, reviewed, documented, and validated by humans.
• Contribute to the technical and strategic direction of the framework and improve quality and abstraction levels across the team.

Skills

Agile / SDLC

IT consulting

LLMs, embeddings, RAG

AI safety & guardrails

Communication & training

Tools

Kiro CLI

GitHub Copilot

Claude Code

Cursor

AWS Bedrock AgentCore

Microsoft Foundry Agents Service

Microsoft Copilot Studio

Job description

Tasks:

• Design and implement reusable framework components for agentic AI systems, including schemas, contracts, automated tests, and documentation
• Create Architecture Decision Records and arc42 documentation, including rejected alternatives
• Establish evaluation procedures for agentic workflows, including test cases, quality metrics, comparison datasets, and reassessment after model changes
• Implement blocking quality gates in GitLab CI/CD pipelines
• Maintain a versioned and searchable component registry and define a contribution process for other teams
• Assess teams’ enablement needs and provide training, workshops, guided use‑case sessions, hackathons, office hours, and demonstrations
• Train teams in the tool stack, AI tools, agentic working methods, quality assurance, limitations, and common AI failure modes
• Create and maintain training materials, guidelines, reusable examples, and reference cases
• Support teams until they can use the framework independently and act as a contact for user questions
• Evaluate new tools, models, and methods through limited trials with documented results
• Coordinate with Architecture, Security, Data Protection, Compliance, and Operations
• Participate in backlog refinements and ART planning, including effort estimation and task slicing
• Ensure compatibility across macOS, Linux, and Windows environments
• Monitor the costs and runtimes of agentic workflows and recommend suitable alternatives
• Define how AI-generated changes are created, reviewed, documented, and validated by humans
• Contribute to the technical and strategic direction of the framework and improve quality and abstraction levels across the team

Required skills:

Must haves:

• Very good understanding of agile working methods and the end-to-end software development lifecycle
• Practical experience with MCP servers and agentic skills
• At least 12 consecutive months of hands‑on experience using AI tools in software development, such as Kiro CLI, GitHub Copilot, Claude Code, Cursor, or equivalent
• Several years of experience in IT consulting and knowledge transfer
• Strong understanding of LLMs, context windows, embeddings, RAG, tool calling, agents, orchestration, guardrails, prompt injection, evaluation, hallucinations, and token costs
• Ability to identify AI limitations and failure modes and derive appropriate implementation decisions
• Ability to present a case where an AI-supported approach was unsuitable and explain the resulting decision
• Ability to explain technical solutions so that other teams can use them independently
• Practical Scrum experience, including backlog work, vertical task slicing, Definition of Done, and Jira
• Ability to design reusable abstractions instead of isolated solutions
• Strong user focus when designing framework components
• Reliable delivery in an agile team and the ability to break down complex initiatives into implementable steps
• Constructive, evidence-based decision‑making and the ability to work effectively with incomplete information
• Proactive learning mindset and willingness to share knowledge
• Strong communication, moderation, analytical, teamwork, and assertiveness skills

Nice to haves:

• Experience in large, regulated IT organizations
• Experience improving agentic systems through evaluations
• Cloud experience, preferably in an organization using the SAFe framework
• Experience defining quality metrics and comparison datasets for agentic workflows
• Knowledge of AWS Bedrock AgentCore, Microsoft Foundry Agents Service, and Microsoft Copilot Studio
• Experience in the railway sector or a comparably complex corporate environment with Security, Data Protection, and Compliance requirements
• Previous experience within the Deutsche Bahn Group