EngineeringJakarta, Indonesia, JakartaFull-time

AI Application Engineer

About the role

This is not a research role and it is not model training. You build the layer that turns a language model into something a business can actually rely on: retrieval over a client's own documents, agents that complete a task end to end, evaluations that prove the thing works, and the guardrails that decide what a model may assume and what has to come back to a human. We focus on four models: Claude, OpenAI, Gemini and Mistral. Which one fits a given task is something we keep testing rather than assume, so building up knowledge of how they differ is part of the job and not an extra.

We are honest about why this seat exists. The hard part of applied AI is not the prompt, it is everything around it: context, boundaries, failure handling, and a repeatable process that still behaves on its hundredth run. If that problem is the one you find interesting, this is a good place to solve it with real clients rather than in a demo.

What you'll do

  • Build retrieval pipelines over client data: chunking, embeddings, vector search, and the evaluation that proves retrieval actually improved the answer.
  • Design and ship agent workflows that complete real tasks, including what happens when a step fails.
  • Define per task what a model may decide on its own and what must return to a human, and make that boundary visible in the system, not implicit in a prompt.
  • Build evaluation harnesses so a change to a prompt or a model is a measured decision rather than a hunch.
  • Integrate models into existing client software through the Vercel AI SDK and direct provider APIs.
  • Choose between the four models on cost, latency and quality, and keep that choice evidence-based rather than habitual.

How we build

We care more about the foundation of a solution than about the speed of the first demo. That means data models and architecture that still hold up in three years. It also means governance, compliance and traceability are designed in from the start instead of bolted on afterwards. When we put agents to work, we record what they did and why, so a decision can always be reconstructed and a client can always be shown the trail. Anyone can get a prototype running. Building something a business can still rely on in two years is the actual job.

Concretely, and starting from the client rather than from us: what we build on depends on the stack they already run and on what will integrate cleanly with it. Where we are free to choose, the default is a TypeScript and Node.js application layer, with the Vercel AI SDK and direct provider APIs on top of Claude, OpenAI, Gemini and Mistral. Azure AI Foundry is what we build business and enterprise applications on, because that is where governance and stability have to be guaranteed rather than argued for afterwards. Python turns up wherever a data or evaluation job is better served by it, PostgreSQL with vector search sits behind retrieval, and React with TypeScript and Tailwind covers the front end. Automation that does not deserve a codebase runs on n8n. You will not have touched all of it, nobody here has, and part of the job is judging when our default is the wrong answer for a particular client.

How we work

You work in one team with our consultants in Eindhoven, not as a separate supplier taking orders. We keep fixed overlap hours so standups, planning and feedback happen live instead of in a comment thread you read the next morning. There are at most two links between you and the client, which means you hear why something is being built and you can push back when the reasoning does not convince you. Our founders travel to Jakarta regularly, and the team travels to the Netherlands. We hold ourselves to the same standard we sell: our own daily operations get improved alongside client work, and when something we built ourselves stops holding up we rebuild it rather than patch around it.

Bonus points

  • You have taken an AI-supported system to production and know where the naive version breaks. RAG counts, so does an automation that replaced real manual work, or an integration that had to keep running unattended.
  • You have compared models against each other on a real task and can show how you measured it.
  • Experience with fine-tuning, and a clear view of when it is not the answer.
  • You have built evaluation sets for a non-deterministic system.
  • You write about what you learn, internally or publicly.

Growth, and what we offer

Where we are right now, so you can judge what you are joining: the Jakarta hub is expanding, and our clients are in the Benelux and the wider EU. That is where the focus sits. Our Dutch engineers sit with the client and build the work together with the team in Jakarta, on the same day rather than across a handover, because what we are working towards is being the technology partner a client trusts with the things that matter rather than a supplier you send a specification to. Growing into the Indonesian market comes after that, and the first commercial seat in Jakarta is where it starts. You arrive while all of this is still taking shape.

  • A competitive salary benchmarked against the Jakarta market for senior engineering work, reviewed every year.
  • Year one you go deep on our stack and our delivery standards. Year two you lead a workstream or mentor the engineers who join after you. When a lead role opens we look inside the team first.
  • We listen to our OneMinders. If you see a better way of working, you get the room to develop it and the backing to actually roll it out. Your opinion counts here, and this company is shaped by the people in it.
  • An annual learning budget tied to that path, not a perk that nobody claims.
  • Hardware of your choice.
  • Direct access to the founders. There is no layer of account managers between you and a decision.
  • Work that funds something: every project contributes to educational initiatives in Indonesia.
  • The chance to work with, and travel to, the team in the Netherlands.

How we hire

You speak with the people you would actually work with, not a screening layer that passes you on. The process is the same in Eindhoven and in Jakarta, and from your application to an offer takes three to four weeks.

  1. 1

    The founders

    Why us, why you, and whether we mean the same thing.

  2. 2

    The Tech Lead

    The work itself: the substance of the role, and what you would actually be doing in it.

  3. 3

    A teammate

    Someone you would work with daily, not the person who would manage you.

Bring something you built to that second conversation and walk us through it. It is not an extra round and not a puzzle we invented: it is that conversation. A side project, something from work you are allowed to show, a repository you are proud of. If everything you have built sits behind an NDA, tell us and we will find another way.

Requirements

  • 5+ years of software engineering, of which at least one building on language models in a real product.
  • Strong TypeScript and Node.js, because that is where our AI application layer lives, with Python wherever a data or evaluation job is better served by it. The AI layer is never more reliable than the engineering underneath it.
  • Hands-on experience with RAG, agent frameworks, or both, beyond a tutorial project.
  • You can explain why a system produced a given answer, and design so that question stays answerable.
  • Professional English, spoken and written. You work with Dutch colleagues and clients every day, so this is a hard requirement.
  • You experiment on your own initiative and bring back a result with evidence, including when the result is that it did not work.
  • You are amplified by AI and still accountable for what ships. Claude, Opencode and the rest of our AI tooling are part of the daily job here, and so is the other half of it: you read and review every line that goes out under your name, and you can defend it. Code nobody can explain does not pass review, whatever produced it.
  • You are curious beyond your own ticket. You think along on our client solutions and on how we run our own shop, and you say plainly where AI genuinely adds value and where a simpler integration or automation would do the job better.
  • Do not tick every box? Apply anyway if you bring real depth, projects you are proud of and knowledge you can demonstrate. That weighs more heavily with us than an exact number of years.
Posted 9 August 2026
Apply now

Questions about this role?

Talk to us