Senior / Lead Data Scientist

• Several years of industrial ML and data science experience. Real projects, real constraints, real stakeholders.
• Statistical depth you can defend under questioning, without reaching for a hand-wave. You understand uncertainty and how to quantify it. You know what a confidence interval does and does not tell you, and what a p-value actually is - and, more revealingly, what it is not. You can feel the difference between a correlation that means something and one that is an artefact. You know what a hypothesis test assumes, and what quietly breaks when those assumptions don't hold.
• A meticulous, close to pedantic relationship with your own results. A number that looks too good makes you suspicious rather than pleased. You check before you believe, and you check the hardest when the work is your own.
• The reflex to interrogate data before modelling it - to plot it, ask how it was collected, and notice what is missing or wrong before drawing a single conclusion.
• Solid Python and the usual ecosystem (scikit-learn, PyTorch or TensorFlow as the problem needs, git, uv). Comfortable on cloud, AWS in our case. server and server-less.
• The ability to carry a room - explaining technical reality to business stakeholders in a way that lands, and caring whether the customer succeeds rather than whether the ticket is closed.
• Fluent English

At Spryfox we build AI solutions that hold up in the real world, for customers across insurance, finance, healthcare, energy and manufacturing. We like to say that success isn't magic, it's great engineering. Good data science works the same way. It rests on understanding what you are doing and why, all the way down. We are looking for a Senior Data Scientist who is strong on the fundamentals - properly strong, the kind that survives questioning. Plenty of people can point a model at a column, call fit, read off 94% accuracy and feel done. If your immediate response is to question what the number is hiding, we would like to talk to you. What the work looks like Our data is mostly tabular, sometimes text, often time series or signal, now and then images. Whatever the modality, it tends to arrive messy, incomplete, biased in ways nobody warns you about, and sparser than you would like. The algorithm is rarely the hard part. The hard part is understanding the data well enough to know which question is even answerable, and being willing to say so when some are not. You would own projects end to end - framing the problem with the customer, choosing the algorithm, carrying out the analysis, and standing in front of the stakeholders to explain what you found and how much weight it can carry. Our clients are at the heart of how we work. Their problems become our problems. When something is wrong, it doesn't let go until it is fixed. We want the customer to actually succeed and we are looking for someone wired the same way.