Data Scientist, Decision Intelligence

Nexonomy AI; Time; On-site; NOK 900,000 - 1,400,000; Full time

Job summary

Nexonomy AI is seeking a talented machine learning scientist to own and calibrate a multi-model advisory system. You will design how independent models reason over the same decision, surface disagreements, and craft a defensible recommendation with a calibrated confidence vector.

You will build evaluation harnesses, define metrics, and collaborate with engineering to deploy the system from notebook to production, ensuring rigorous uncertainty quantification across high-stakes decisions.

Qualifications

• Applied ML or statistics background with model deployment experience.
• Fluency with modern LLMs and ensemble methods.
• Rigorous handling of uncertainty and high-stakes decisions.

Responsibilities

• Design and evaluate the multi-model advisory: how independent models reason over the same decision, where they agree, and how disagreement is surfaced rather than averaged away.
• Own the calibration behind our confidence vector and evaluation rubric, so a recommendation's confidence means something and holds up a year later.
• Build the evaluation harnesses and metrics that measure decision quality, and re-fit them as frontier models change.
• Work with engineering to take this from notebook to in-tenant production.

Skills

Applied ML

Statistics

Model deployment

Model evaluation

LLMs ensemble

Uncertainty

Tools

Python

PyTorch

Experiment tracking

Job description

The decision system of record is only as trustworthy as the engine behind it. You will own the science that turns several models' opinions into one calibrated, defensible recommendation, the part of Nexonomy a fast follower cannot copy.

What you will do

• Design and evaluate the multi-model advisory: how independent models reason over the same decision, where they agree, and how disagreement is surfaced rather than averaged away.
• Own the calibration behind our confidence vector and evaluation rubric, so a recommendation's confidence means something and holds up a year later.
• Build the evaluation harnesses and metrics that measure decision quality, and re-fit them as frontier models change.
• Work with engineering to take this from notebook to in-tenant production.

What we are looking for

• A strong applied ML or statistics background; you have shipped models or evaluation systems that real decisions depended on.
• Fluency with modern LLMs and how to ensemble, evaluate, and calibrate them.
• Rigor and intellectual honesty about uncertainty; we never paper over a weak signal.
• Comfort owning ambiguous, high-stakes problems end to end.