Senior Machine Learning Engineer, Decision Intelligence

Atlas is building a learning decision intelligence system. We already have a chief architect and a small machine learning team, but the work is not standard ML. It is decision intelligence: models that reason about operational state, uncertainty, and consequence, and whose outputs are acted on by institutions. We are looking for engineers who have shipped production ML and want to work at this layer.

• Design and train predictive models for operational forecasting, risk indication, and anomaly detection across domains.
• Fine-tune and adapt foundation models where they add value, and build domain-specific models where they do not.
• Own the model lifecycle end to end: data requirements, training, evaluation, deployment, monitoring, and retraining.
• Work with the chief architect and the intelligence team on how uncertainty is represented, calibrated, and surfaced.
• Collaborate with data and platform engineers on feature pipelines, training infrastructure, and inference serving.
• Write down what you build. Our models have to be explainable to people who did not build them.

• Strong production machine learning experience: you have shipped models that ran in a real system and you owned their behaviour after release.
• Comfort across the model lifecycle, from data preparation through training, evaluation, and serving.
• Working knowledge of probabilistic reasoning and time series methods, and the judgement to know when they apply.
• Ability to write production code, not only notebooks. Python is our working language.
• Ability to explain model behaviour to engineers, domain experts, and clients.

These are not required. They are the kind of background that tends to do well in this role.

• Prior experience with decision-support systems, operational forecasting, or simulation.
• Experience fine-tuning or adapting foundation models for domain-specific tasks.
• Background in energy, logistics, infrastructure, or public-sector systems.
• Familiarity with provenance-aware or audit-oriented machine learning.