AI AI Engineer Hoofddorp · hybrid

Devleaps; Hoofddorp; Hybrid; EUR 70.000 - 100.000; Vollzeit

Zusammenfassung

Devleaps in the Netherlands seeks an Applied AI Engineer who designs and deploys business-ready AI agents, integrating with client systems and ensuring guardrails and observability. You will start from real business problems, build multi-step workflows, handle legacy data, and productionize retrieval and evals.

Fluency in Dutch and professional English is required; experience with Azure, AWS or similar cloud and agent frameworks is a plus.

Qualifikationen

• 4–8 years of software engineering experience, with at least 1–2 years building LLM, agent or generative AI solutions that real users depend on, ideally in production.
• Strong Python or TypeScript skills and experience integrating with enterprise systems through APIs, events and messy data.
• Hands-on experience with LLM APIs and agent frameworks (Anthropic, OpenAI, Azure OpenAI or Bedrock; LangGraph, the OpenAI or Claude Agent SDKs, Semantic Kernel or similar).
• Experience with tool use and MCP, multi-step and multi-agent workflows, structured output and human-in-the-loop patterns.
• Experience building and evaluating RAG: chunking, embeddings, hybrid search, re-ranking, retrieval quality and hallucination mitigation, on vector stores such as pgvector, Azure AI Search or similar.
• Hands-on work with evals, guardrails and LLM observability (tracing, cost and quality metrics), with tools such as Langfuse, LangSmith, Braintrust, orq.ai or similar.
• Working knowledge of identity, authorisation and data protection in the context of agents, and familiarity with what the EU AI Act asks of the systems you build.
• Comfort with CI/CD, containers and at least one cloud (Azure or AWS preferred).
• Experience with an enterprise agent platform such as Copilot Studio, Agentforce, ServiceNow or Gemini Enterprise is a strong plus, as is having scoped or delivered a fixed-outcome project.
• Fluent Dutch, professional English.

Aufgaben

• Start from the business problem, find a measurable use case, and connect it to values like cost to serve, throughput, revenue or customer experience.
• Design and build agents that survive contact with reality: multi-step workflows, tool and API orchestration, structured output and human-in-the-loop when needed.
• Work the messy middle: integrate with legacy systems, handle partial data, identities and permissions, and manage exceptions.
• Make agents safe to grant real access: scope what an agent can see/do, build input/output guardrails, and maintain an audit trail for security/compliance.
• Build retrieval that works on real content: RAG and search pipelines against client documents and policies, measured by retrieval quality.
• Prove it works and keep proving it: write domain evals with the business and set up monitoring for ongoing validity.
• Build on what the client already has: deliver on Azure AI Foundry, Bedrock, Vertex or other platforms and offer honest feedback on gaps.
• Take it to production and hand it over well: ship one live use case with evals, guardrails and documentation, then provide reusable building blocks.

Kenntnisse

Python

TypeScript

LLM APIs

Agent design

API integration

CI/CD

Cloud (Azure/AWS)

Dutch fluency

English proficiency

Tools

Anthropic

OpenAI

Azure OpenAI

Bedrock

LangGraph

Semantic Kernel

Langfuse

LangSmith

orq.ai

Jobbeschreibung

You build AI that the business actually uses, connected to the systems it actually runs on, and you stay until it works.

At Devleaps we help organisations get past the pilot. A lot of companies have run a chatbot experiment, bought an AI platform and seen a demo that impressed the board, and then watched the whole thing stall somewhere between the demo and production. We come in for the part that decides whether it lands: the agent that has to call a fifteen-year-old API, read documents nobody ever cleaned, respect who is allowed to see what, and still give an answer the business can put its name to.

We are looking for an Applied AI Engineer who enjoys exactly that part. You design agents and AI solutions around a real business problem and a number the business already tracks, you build them on whatever AI foundation the client already runs, and you make sure they are evaluated, guarded and observable before anyone relies on them.

WHAT YOU'LL DO

• Start from the problem, not the model: sit with the business owner, find the use case that is worth solving, and connect it to something they already measure, such as cost to serve, throughput, revenue or customer experience.
• Design and build agents that survive contact with reality: multi-step workflows, tool and API orchestration, structured output and human-in-the-loop where the stakes require it, so the agent does useful work instead of only chatting.
• Work the messy middle: integrate with legacy systems, partial data, existing identity and permissions, and all the exceptions nobody wrote down, because that is where most AI projects quietly die.
• Make agents safe to give real access: scope what an agent is allowed to see and do, build input and output guardrails, and leave an audit trail that security and compliance can sign off on.
• Build retrieval that works on real content: RAG and search pipelines against the client's actual documents, policies and records, measured on retrieval quality rather than on a clean demo corpus.
• Prove it works, and keep proving it: write domain evals with the business so everyone knows the solution is good before it ships, and set up the monitoring that shows it is still good three months later.
• Build on what the client already has: deliver on Azure AI Foundry, Amazon Bedrock, Google Vertex, Gemini Enterprise, orq.ai or whatever the organisation already bought, and give an honest opinion on what is missing rather than a migration pitch.
• Take it to production and hand it over well: ship one use case live, with evals, guardrails and documentation in place, and leave reusable building blocks behind so the next use case goes faster, whether we build it or the client's own team does.

YOU ARE

• An engineer first: you care about system design, reliability, integration and maintainability, and AI is the domain rather than a reason to skip the fundamentals.
• Business-minded: you can talk to a business line owner about their process and their numbers, and you would rather ship one agent that moves a metric than five that impress in a demo.
• Foundation-agnostic: you pick the model, framework and platform that fit the client, and you are comfortable building inside someone else's stack.
• Trustworthy with real access: you think about permissions, data protection, failure modes and cost per call before anyone asks, because an agent that touches production systems has to be boring in all the right ways.
• A bridge-builder: you work easily with business owners, data engineers, security teams and the client's own developers, and you help them understand each other.
• AI-native in how you work: you use coding agents every day and know from experience where they help and where they fall over.
• Curious and honest: the field moves monthly, you learn fast, you know what you know, and you tell a client when the right answer is a rules engine or a better process instead of an LLM.

YOU'LL BRING

• 4–8 years of software engineering experience, with at least 1–2 years building LLM, agent or generative AI solutions that real users depend on, ideally in production.
• Strong Python or TypeScript, and solid experience integrating with enterprise systems through APIs, events and messy data.
• Hands-on experience with LLM APIs and agent frameworks (Anthropic, OpenAI, Azure OpenAI or Bedrock; LangGraph, the OpenAI or Claude Agent SDKs, Semantic Kernel or similar).
• Experience with tool use and MCP, multi-step and multi-agent workflows, structured output and human-in-the-loop patterns.
• Experience building and evaluating RAG: chunking, embeddings, hybrid search, re-ranking, retrieval quality and hallucination mitigation, on vector stores such as pgvector, Azure AI Search or similar.
• Hands-on work with evals, guardrails and LLM observability (tracing, cost and quality metrics), with tools such as Langfuse, LangSmith, Braintrust, orq.ai or similar.
• Working knowledge of identity, authorisation and data protection in the context of agents, and familiarity with what the EU AI Act asks of the systems you build.
• Comfort with CI/CD, containers and at least one cloud (Azure or AWS preferred).
• Experience with an enterprise agent platform such as Copilot Studio, Agentforce, ServiceNow or Gemini Enterprise is a strong plus, as is having scoped or delivered a fixed-outcome project.
• Fluent Dutch, professional English.