AI engineer

An opportunity for an experienced AI Engineer to join a leading financial institution in Amsterdam and help accelerate the adoption of AI across the business.

You will build modern, production-grade AI systems using agents, retrieval, memory and evaluation loops. Working at the intersection of business and engineering, you will design and deliver hands-on AI applications that solve real business challenges. The environment runs fully on AWS, and the work is focused on application building rather than low-code solutions.

Responsibilities:

• Design and implement agent workflows, including context engineering, tool use and orchestration.
• Build retrieval and memory systems, including vector search, session memory, long-term memory and compaction.
• Develop evaluation pipelines, including automated evaluation, human evaluation and evaluation suites.
• Integrate AI services on cloud platforms, particularly AWS Bedrock, Knowledge Bases and agent frameworks.
• Build AI use cases for the business, with around 20 use cases currently in the pipeline.
• Work closely with business stakeholders to design fit-for-purpose AI solutions.
• Deliver production-grade AI applications using modern software engineering practices.

Requirements:

• Strong AI/ML background with hands-on experience building AI solutions.
• Experience setting up and delivering AI use cases for business stakeholders.
• Hands-on experience with LLMs through APIs or SDKs.
• Experience with agent-based systems, including memory, orchestration and multi-step workflows.
• Strong knowledge of retrieval systems, embeddings and vector search.
• Experience with evaluation frameworks and quality measurement for AI systems.
• Experience working with AI solutions on AWS.
• Experience with AWS Bedrock is required.
• Experience with AWS AgentCore is a strong advantage.
• Strong software engineering background, including Python and version control.
• Experience with TypeScript.
• A hands-on engineering mindset and the ability to build real applications rather than relying on low-code tooling.