Senior Data Scientist / ML Engineer

Kyper is the agentic downtime partner for heavy industry. Mining and oil & gas operators lose millions every hour to unplanned downtime. Our ML brain catches anomalies before alarms fire, assembles the full context behind them, recommends the next move, and gets sharper every cycle - identify, diagnose, recommend, learn. Live with customers. Scaling fast.

Don't apply if

• You can only work from a clean, fully written spec.

• You'd rather not talk to customers directly.

• Your data science experience is mostly notebooks, coursework, or research papers.

• You have never owned a model in production end-to-end.

• Your AI tooling stops at chat interfaces and basic IDE autocomplete.

• You need a fixed 9-to-5. Customers span Japan to Alaska, so we trade strict hours for flexibility on both ends.

• You want every part of the role mapped out before you start. We are still building, and priorities shift with what customers and the product need.

The role

You walk into a customer call with a vague problem and walk out with a sharp experimental plan. You own data science work end-to-end: scoping with customers, designing experiments, picking tooling, training models, shipping them to production, and iterating on what the data tells you.

Half this job is figuring out which question to ask. The other half is shipping the answer.

Our customers are world-class at heavy industry. They know their equipment, their failure modes, and their operations better than anyone in the world. What they typically do not have is deep data science or software depth. The knowledge you need to build good models will not arrive on a silver platter. You will extract it - through conversations, observation, asking better questions, and being creative about what to test. If you need a domain expert sitting next to you translating every signal, this is not the right job.

What you'll do

• Design and run experiments end-to-end: data exploration, feature work, model selection, evaluation, calibration.

• Ship models to production on Vertex AI, SageMaker, or equivalent, and keep them healthy.

• Own the tooling decisions: tracking, registry, pipelines, serving. Pick what fits and defend the choice.

• Sit with customers, understand the operational problem, and translate it into testable hypotheses.

• Push results back to the customer, learn from what they tell you, iterate fast

You bring

• 5+ years of commercial data science / ML experience, with real models in real production environments. Not notebooks, not Kaggle.

• Strong Python, PyTorch (or equivalent), MLflow or similar tracking, and shipping experience on at least one cloud ML platform (Vertex AI, SageMaker, Azure ML).

• Opinions about tooling, and the receipts to back them up.

• Comfort with ambiguous problems and direct customer interaction. You do not need a perfect spec to start.

• A track record of taking ideas from POC all the way to production. This is non-negotiable.

Bonus

• Time series, anomaly detection, predictive maintenance, or industrial / IoT data.

• Distributed training, Ray, or Spark at scale.

• Built or designed evaluation harnesses for production ML.

• PhD in a practical, applied field: anomaly detection or related ML, applied physics (not theoretical), or a hands-on engineering discipline. Pure-theory PhDs do not carry the same weight here.

You will fit Kyper if you

• Are productive with AI code agents and modern AI-assisted development workflows. If you have never built a Claude Code skill or subagent, set up an MCP server, run a personal knowledge system like Obsidian, or chained AI tools into a real workflow, your AI usage is probably too basic for what we need.

• Move fast, learn fast, and adapt quickly.

• Can take ideas and POCs all the way to production. This is a must-have skill.

• Are willing to dive into whatever challenges are needed to get things done.

How we work

Small team, real customers, real impact. Lots of ownership, no fixed playbook, plenty of room to set your own approach. We hire people who want to be judged by outcomes.

Fully remote on a B2B contract. Quarterly team co-locations in Warsaw, plus occasional travel for customer visits. Customers span Japan to Alaska, so the occasional early or late call comes with the role. The trade is real flexibility on the rest of your time.