Artificial Intelligence Consultant

We are supporting a growing AI product company that is developing an enterprise AI platform for serious production environments.

They are looking for a hands-on AI Architect who can shape the technical direction of the AI platform while remaining close enough to the engineering to validate architectural decisions through code.

This is not a purely advisory or presentation-led architecture position. You will make practical technical decisions based on previous experience of designing, deploying and operating AI systems in production.

Your responsibilities

• Define the architecture for production AI agents and applications
• Shape multi-tenant RAG architecture, retrieval strategy and data isolation
• Design vector search, hybrid retrieval, re-ranking and index lifecycle processes
• Establish evaluation, observability and release standards
• Design for explainability, audit lineage, decision provenance and security
• Define model access, routing, tool integration and tenant-level usage controls
• Make architecture decisions covering quality, latency, token usage, cost and reliability
• Align the AI layer with APIs, service connectivity and Kubernetes infrastructure
• Define Model Context Protocol integrations with enterprise tools and services
• Produce architecture decision records, technical standards and operational runbooks
• Build or review reference implementations for critical parts of the platform
• Support engineers with technical direction, design reviews and production problems

Must-have technical experience

Applicants must be able to demonstrate hands-on production and architectural experience across the following technology environment:

• Python
• FastAPI
• PydanticAI
• LangGraph
• LiteLLM
• Langfuse
• PostgreSQL and pgvector
• Model Context Protocol
• Kubernetes
• Production AI agents and agentic applications
• Multi-tenant RAG architecture
• Vector search, hybrid retrieval and re-ranking
• Index design and lifecycle management
• LLM and retrieval evaluation frameworks
• LLM observability and production monitoring
• API and enterprise service integration
• CI/CD, automated testing and controlled production releases
• Tenant isolation, access controls and usage quotas
• Monitoring latency, token consumption, cost, failures and retrieval quality
• Explainability, audit trails, decision provenance and operational traceability
• Secure AI systems within regulated or highly governed environments

What we are looking for

• Substantial professional Python and AI engineering experience
• A proven record of designing, deploying and operating AI or LLM systems in production
• Experience making architectural decisions for enterprise AI platforms
• Strong understanding of production reliability, failure modes and system performance
• Enough hands-on ability to validate architectural decisions through code
• Experience documenting technical decisions, trade-offs and operational procedures
• The ability to explain how systems performed under load and where they failed
• Experience improving architecture based on incidents, evaluation results and user behaviour
• Residence and work authorisation in Germany
• Willingness to attend occasional team meetings in Munich

Experience within banking, insurance, pharmaceuticals, healthcare or another regulated environment would be particularly relevant.

What you can expect

• Salary of up to €130,000, depending on experience
• Virtual Stock Option Plan participation
• Fully remote working within Germany
• 30 days' annual leave
• A choice of Edenred meal and shopping vouchers or EGYM Wellpass
• Significant influence over the platform's technical direction
• Architectural responsibility from your first project
• Direct access to the leadership and AI teams
• Professional exchange with highly experienced AI specialists
• A modern technology environment with room to test and evaluate new approaches
• Occasional team meetings in Munich

When applying, please include a short and specific answer to this question:

What AI system have you personally implemented into production, how long has it been running and what is its biggest weakness today?