The posting
We're looking for one data engineer who loves getting under the skin of data, and who is excited about what agents now make possible.
Agents have made data-driven development possible where it wasn't before. Messy real-world data needs ground truth and evaluations, and building them used to be very expensive. Now evaluation harnesses, with agents developing against them in a loop, can take the processing of such data to a precision and recall that wasn't possible before. What this needs is someone who understands what the data means, models it properly, and builds the harnesses that tell agents what right looks like.
IPRally is building a new generation of AI powered tools for understanding patents and technical knowledge, and everything our product does depends on patent data. Offices around the world publish it, each in its own format and language, and some of it exists only as scans of paper. Today it reaches us through data providers, and we expect to get more of it straight from the source.
Our users should see the right data in the right way. Your part of that is getting the data right and delivering it into the stores and APIs our product is built on. From there, product engineers, designers and product managers turn it into what users see. Doing your part well takes taste, and a feel for how the whole system fits together, from the sources to the product.
You'll join our data engineering team of three engineers and work closely with the other engineering teams and the rest of IPRally.
What you'll do
You'll learn what our data means, find where it falls short, and model it so that it reflects how patents, research papers and the companies behind them relate to each other.
You'll build the harness and evaluations that let agents fix data problems on their own, and judge when their results are good enough.
You'll build and run the pipelines that take data from our sources into the stores and APIs the product reads from. They're written in Python, run with Dagster, and monitored with data quality checks that show when something has gone wrong.
You'll work with the product teams on what data they need and in what form, and with the ML team on how we use machine learning in our processing steps.
You'll explore what new sources could give us, including data straight from patent offices, and help decide what's worth adding. With agents, we can try out a new source quickly before committing to it.
When a source stops delivering or falls short, you'll find out why and help work out what other sources could fill the gap.
What we're looking for
This role could be a strong fit if you:
- Have built and run production pipelines on large, noisy real-world data, or worked with such data in statistics or ML.
- Are comfortable deciding what to build when there's no fixed definition of correct and every tool is imperfect, and build statistical evaluations and measurable targets into your work from the start.
- Build agent harnesses that develop our data processing systems on the same basis, so that we can be confident in the quality of both the data and the code.
- Are curious about the real world behind the data: how it came to be, what it means, and how to represent it in the systems that process it.
- Can work out what right data looks like when everything around it is imperfect, from who uses it, what they use it for.
- Like working directly with the people who use the data, inside and outside engineering, to decide what's worth doing.
Experience with patent or other document data, distributed processing such as Ray, or running ML models over large datasets is an advantage, but isn't required.
How we build
Our engineers aren't measured by how much code they personally type. We're much more interested in whether you can understand the data and what people need from it, make good decisions, and use the tools available to turn those decisions into reliable pipelines.
AI agents are already a normal part of how we build at IPRally, and we're still working out what they change in data work. You'll have considerable room to experiment and help shape how the data team works with agents.
Our data stack includes Python, Dagster, Ray, Polars, Postgres, Vespa, Kubernetes and GCP. You don't need to know all of it, or anything about patents: we're much more interested in your judgment and ability to learn.
The salary range for the role is €5,500–7,000/month, with equity on top.
Ideally you're based in Helsinki, but elsewhere in Europe can work for the right person.
Along with your CV, write half a page answering this question: What's the most interesting dataset you've worked with, and what did it take before you could trust it? There's no right answer. We want to see how you think.
