Data scientist / ML engineer

Ship maintenance is broken. We're fixing it.

Ceto is a fleet intelligence company. Our platform, Watchkeeper, takes the high-frequency sensor data ships already generate and predicts machinery failures before they happen — on one vessel we flagged a developing fuel-system issue three days before a critical failure, on a ship earning $48,000 a day. No new hardware, no rip-and-replace: if an owner is already capturing data, we can have a full fleet live in days. We're deployed across fleets owned by major global shipowners, and we sell directly to the ship owners and operators who feel every day of unplanned off-hire on their P&L.

We're a small, senior team - our CEO spent eight years at sea as a marine engineer before founding the company; we’re backed by investors including Dynamo Ventures who led our $4.8m seed round, and we're compounding fast which is exactly why we need you.

What we're doing

We're reshaping maritime. It's an industry that's incredibly important and serves as the backbone of global trade, moving $14 trillion of cargo each year. Despite its significance, the industry remains one of the final frontiers of digitalisation.

Ceto provides the operational intelligence layer that technical teams rely on for mission-critical decision making. By capturing and analysing high-frequency data from commercial ships, Ceto monitors the condition of the machinery onboard. Through anomaly detection and degradation forecasting we predict mechanical failures ahead of time and guide decision making for our customers. Our analytics are served as insights through our SaaS platform, preventing costly breakdowns, loss of hire events and reducing operational risk.

The role

You'll turn high-frequency sensor data from ships' machinery into analytics our customers act on: condition monitoring, performance baselines, degradation trends, anomaly detection. Some of this is machine learning; much of it is statistics, signal processing and a real understanding of the physical systems producing the data. You'll work closely with the engineering team to take problems end to end, from raw, messy sensor data through to a deployed, monitored output feeding our dashboard.

Your responsibilities will include

•  Building time-series analytics on machinery sensor data: performance baselines, derived signals, trend and degradation detection
• Owning the data from the point it arrives: ingesting raw sensor feeds, cleaning, validating and resampling them, and building the checks that catch gaps, outliers, sensor faults and vendor tag differences before they reach any model or customer
• Exploring new ideas in predictive maintenance and condition monitoring, and turning them into reliable, production-ready features
• Applying ML, supervised or unsupervised, where the problem calls for it, including cases where failure examples are rare and labels are imperfect
• Building and maintaining analytics pipelines end to end (cleaning, feature engineering, model training and inference, output storage) using version control, testing, CI/CD and experiment tracking
• Working closely with the rest of the technical team to prioritise what's worth investigating and what isn't
• Using AI tools such as Claude Code to work faster and more effectively

Two firm requirements

• You have experience in an early-stage startup and are comfortable working at speed, context switching and turning ambiguous problems into delivered solutions
• You’re AI-native, using the modern tools to increase your productivity and you have a good feel for when to trust agents and when to do the work by hand

You'll be perfect for this role if

• You have strong Python skills and know its data/scientific libraries (pandas/numpy/scipy, scikit-learn, statsmodels)
• Experience with supervised and unsupervised ML (e.g. gradient boosting, clustering, hidden Markov models), and clear reasoning about when you'd use each
• You have a strong grasp of time-series analysis (trends, seasonality, filtering, changepoints, anomaly detection) and the judgement to pick the simplest method that works
• You've shipped analytics or models into production, not just notebooks, with exposure to CI/CD and DevOps practices
• You have a physics, engineering or other quantitative degree, and a genuine interest in the physical systems behind the data, not just the data itself
• You can explain why a method fits a physical problem, and you're sceptical by nature. Models can be wrong, intuition can be disproven, and correlations in data often have mundane explanations. You want to understand why, not just what
• You frame problems clearly before solving them, want to build something new, and are comfortable with the pace and ambiguity of a startup
• You're a team player who is confident enough to challenge ideas and put forward your own

It's also a bonus if you have

• Experience in predictive maintenance, condition monitoring, or anomaly detection on sensor or telemetry data
• Experience with cloud platforms (ideally Azure, otherwise AWS or GCP), NoSQL databases such as MongoDB, or ML lifecycle tooling such as MLflow

We're less interested in ticking off niche experience like maritime or MongoDB specifically, and far more interested in someone who has worked across different technical stacks and can pick up new ones quickly. A strong generalist who's genuinely curious about the wider engineering problem, not just the modelling, will do better here than a narrow specialist.

What's on offer

• Competitive salary based on experience
• Equity options
• 25 days holiday plus public holidays
• Birthdays off
• Bupa medical insurance
• Medicash
• Cycle to work scheme
• Hybrid working from our London office (3 days in office)
• Budget and time allocation for conferences and professional development
• The opportunity to shape the future of maritime technology

The interview process

1. Intro call + past project review + take home task (video call)
2. Take home task review + Q&A (video call)
3. Final in person round (London)