We built our own AI recruitment tool, and learned a lot about agentic systems along the way

When we hire at SkyeTec, applications now run through a tool we built ourselves. It is a recruitment system (ATS) that receives applications from skyetec.ai, helps us assess candidates with AI, and keeps the process on track all the way to a signed contract.

We could have bought an off-the-shelf solution. We chose to build our own, and it was a deliberate choice.

Why build it ourselves?

Two reasons.

First, recruitment is an area where the details matter. How a candidate is assessed, which requirements weigh most, how long data is kept, who makes the decision. These are things we want to own ourselves, not inherit from a vendor's assumptions.

Second, and just as important: we make our living building AI solutions for others. So we have to build them for ourselves too. This project is dogfooding put into practice. We apply our own methods and tools to a real problem with real consequences, so that what we recommend to our clients is something we have actually lived with ourselves.

What we built

A complete recruitment system that covers the whole journey.

Application intake straight from our career pages, with job postings managed from the system itself.

AI-assisted candidate assessment, where the model reads the CV and application, extracts structured facts, and the system computes a score against the requirement profile for the role. The recruiter gets a candidate card with a summary, observations and a reasoned score, as a starting point, not a verdict.

An interview pipeline from the first meeting to the technical round, with a task library and structured feedback between rounds.

Automatic handover to HR. When someone is hired, the employee and the contract are created in our HR system (Huma) automatically.

Privacy built in from the ground up. Deletion deadlines are set on intake and adjusted as the process moves along, and personal data is removed from the text before it is sent to the AI model.

Everything runs in our own cloud infrastructure on Azure, with data and AI processing kept inside the EU.

How we built it, and why that is the real point

This is what we are most proud of, and what we carry into client projects.

We built it agentically. The system is not just an app with an AI button. The AI does a bounded, verifiable piece of work, reading and structuring information, while the scoring itself happens in ordinary, deterministic code. The same CV produces the same score, every time, and we can always explain why. It is an architecture we believe strongly in. Use the language model for what it is good at, and let the code handle what has to be predictable and traceable.

We also went a step further and gave the system its own agent interface, an MCP server, so that AI agents can work against the recruitment pipeline in a controlled way. They can read status, create roles and register candidates, behind proper authentication. It is the same kind of agentic integration we build for clients, tried out on our own house first.

And we built it with AI close to the development itself. All the way from specification to code, infrastructure and documentation, we worked AI-centrically. Decisions are discussed and documented, the code is written in close interplay between developer and AI tooling, and lessons are written down as we go so they become reusable rather than forgotten. The result is a small team that delivered a full system quickly, without compromising on traceability and quality.

Responsible AI is not an afterthought. AI in recruitment is high risk under EU regulation, and that shaped the design. A human always makes the final decision, the model never sees personal data it does not need, and the version of the rules an assessment was made under is traceable. This is how we believe AI should be put into production, even when no one is forcing us.

What we take with us

What we are left with is not just a recruitment tool. It is a way of working. How to build agentic applications that are fast to make, safe to live with, and that actually hold up when they meet reality, with personal data, regulation and people who have to make decisions.

That is the experience we bring into the projects we do for you.