Applied Scientist, Causal Inference

Octaipipe; City Of London; On-site; GBP 110,000 - 150,000; Full time

Job summary

OctaiPipe is seeking an Applied Scientist in causal inference to own the scientific reasoning behind our data-centre decision intelligence product. You will work on root-cause inference, counterfactual evaluation, and transferable causal models across sites to reduce maintenance effort and downtime.

Collaborating with reinforcement learning scientists and ML systems engineers, you will move methods from prototype to production, with rigorous evaluation, drift monitoring, and clear communication

Qualifications

• Master’s degree or higher in statistics, ML, econometrics or related quantitative field focused on causal inference.
• Hands-on experience applying causal methods to real data: structural causal models, causal discovery, quasi-experimental designs, or counterfactual estimation.
• Strong time-series skills, including data with gaps, drift and regime changes.
• Experience with probabilistic modelling and uncertainty quantification, including calibration of confidence.
• Strong instinct for identification: spotting natural experiments and designing experiments.

Responsibilities

• Own root cause inference: explain departures from expected-healthy behaviour to identify one cause per issue from alarms, telemetry and topology.
• Develop methods that separate genuine plant faults from sensor faults, known defects, expected control actions and planned work.
• Guarantee calibrated confidence and cited evidence on every diagnosis for operators.
• Own model validity: detect when a question falls outside the operating regime and communicate uncertainty.
• Build prioritisation that improves with understanding.
• Estimate treatment effects via counterfactuals and natural experiments.
• Learn how faults unfold from historical incidents across sites.
• Refine the causal structure when data shows relationships are wrong.
• Develop causal models that transfer between heterogeneous sites in collaboration with RL scientists and ML systems engineers.
• Take methods from prototype to production with evaluation, staged rollout and monitoring for drift.
• Raise scientific rigor through evaluation, code review, and clear communication.

Skills

Causal inference

Time-series analysis

Python

Probabilistic modelling

Model deployment

Education

Master's degree or higher in statistics, machine learning, econometrics or related quantitative field

Tools

Python

NumPy

pandas

PyTorch

Job description

Applied Scientist, Causal Inference

Transferable Intelligence Team

The Company

OctaiPipe is a young, ambitious company with the vision to be the global driving force for the next paradigm of foundational, physical AI that ensures our connected world, and its critical infrastructure, is safe, secure and sustainable. We are growing fast, having closed a recent funding round and looking to accelerate rapidly. OctaiPipe is offering the right candidate an exciting role on this adventure!

OctaiPipe is on a mission to revolutionise the optimisation of energy in data centres through decentralised artificial intelligence (AI). To do this, OctaiPipe is harnessing an elegant but revolutionary idea. Rather than move the data from the source, move the algorithms to the data to learn at the data source. This learning can be achieved with the intelligence of many devices through novel federated AI technology. OctaiPipe is developing the AI for Cooling Efficiency (ACE) application to be deployed using its own in-house distributed AI platform.

The Role

The Transferable Intelligence team builds the capability that lets what OctaiPipe's AI learns at one site benefit others: reinforcement learning agents for control, and causal inference and root cause analysis models, that train at the edge and transfer between heterogeneous data centre environments.

We are looking for an Applied Scientist in causal inference to own the causal reasoning behind our decision intelligence product for data centre operators. The product shows operators what needs attention on site, why it is happening and what to do about it.

This demands causal, not correlational, thinking. Your methods must find the cause, say how confident they are, and be checked against what actually happened.

The role also has a transfer dimension. Sites differ in topology, equipment and control strategy, and you will develop causal models that carry across the estate, so what is learnt at one site reduces the data and effort needed at the next.

Duties and Responsibilities

• Own root cause inference: explain departures from expected-healthy behaviour to find one cause per issue from alarms, telemetry and topology.
• Develop methods that separate genuine plant faults from sensor faults, known defects, expected control actions and planned work, in closed-loop plant.
• Guarantee calibrated confidence and cited evidence on every diagnosis, so operators can see why the product reached a view.
• Own model validity: detect when a question falls outside the operating regime a model has seen, propagate that into confidence, and allow the product to say it doesn't know rather than extrapolate erroneously.
• Build prioritisation that improves with understanding.
• Estimate the effect of treatments through counterfactual comparison.
• Find and exploit the natural experiments.
• Learn how faults unfold from historical incidents, pooled across sites where faults are rare at any one.
• Refine the causal structure of the system where data shows documented relationships are wrong.
• Develop causal models that transfer between heterogeneous sites, in close collaboration with the team's reinforcement learning scientists and ML systems engineers.
• Take methods from prototype to production with the ML systems engineers: evaluation on historical replay, staged rollout and monitoring for drift.
• Raise the scientific bar through rigorous evaluation, code review, documentation and clear communication of results to engineers and operators.

Your Profile

• A master's degree or higher in statistics, machine learning, econometrics, epidemiology or a related quantitative field, with a focus on causal inference.
• Hands-on experience applying causal methods to real, messy data: structural causal models, causal discovery, quasi-experimental designs (e.g. difference-in-differences, instrumental variables, regression discontinuity, synthetic control), or counterfactual and treatment-effect estimation.
• Strong time-series skills, including data with gaps, drift and changes of regime.
• Experience with probabilistic modelling and uncertainty quantification, including checking that confidence is calibrated.
• A strong instinct for identification: spotting natural experiments in observational data and designing experiments that answer the question.
• Excellent understanding of Python and the scientific stack (e.g. NumPy, pandas, PyTorch), with the engineering discipline to take methods from prototype into production using modern LLM-based development tools.

You Also Might Have

• Experience with causal inference on graphs or networks, or graph neural networks.
• Experience with physical or engineered systems: data centres, HVAC, power systems, process plants or other industrial settings.
• Fault diagnosis, root cause analysis, or prognostics and health management.
• Hybrid methods that combine physics-based models or simulation with machine learning.
• Causal effect estimation in closed-loop or controlled systems.
• Transfer learning, domain adaptation or federated learning across heterogeneous environments.
• Learning from human feedback or weak labels.
• Publications or open-source contributions in causal inference.

Why Join OctaiPipe

• Work on real-world sustainability impact at global scale
• Influence how AI is responsibly applied to critical infrastructure
• Join a well-funded, rapidly growing scale-up with ambitious goals
• Collaborate with experts across AI, infrastructure, and operations
• Shape a product that can materially reduce energy use and carbon emissions worldwide

The above statements are not intended to encompass all functions and qualifications of the position; rather, they are intended to provide a general framework of the requirements of the position. Job incumbents may be required to perform other functions not specifically addressed in this description.