Oreum Industries is a data science consultancy from Jonathan Sedar in collaboration with a worldwide network of expert developers and leaders. We focus on the insurance industry, helping underwriters, insurtechs and corporates to learn from data. We operate remotely & globally, with a client base spanning USA, UK, Europe and the Middle East.

Our technical bread and butter is advanced Bayesian statistical modeling, to create an edge in pricing & reserving throughout general & life insurance. This is essential to quantify uncertainty and support underwriter-led real-world decision-making in low data environments.

As opportunity allows, we host a handful of technical resources here. In each case at least a Technical Overview is public, and in some cases we've made the code available to read, or even fully open-source. Naturally we can't open everything, so where code is closed, please contact [email protected] for commercial interest - we'd be more than happy to discuss in detail.

These are the result of a concise exploration and implementation of specific tools & techniques relevant to our modeling work. They often contain novel model architectures, describe theory and practice at a deep level, and use Bayesian inference and a Bayesian workflow, specifically using the pymc & arviz ecosystem.

• Oreum Copula Copula Regression Modeling. This provides an extremely valuable method to estimate Expected Loss Costs when Claims Frequency & Severity are correlated in low data environments. Demonstrated 32 percentage-point improvement in accuracy vs non-copula model. Publicly Readable Code, Commercial License

• Oreum Survival Survival Regression Modeling. This provides an extremely valuable suite of model architectures to estimate time-to-event via right-censored Accelerated Failure Time (AFT) and semi-parametric (CoxPH) models. Publicly Readable Code, Commercial License

• PYMC Examples: GLM-ordinal-features Full Bayesian workflow example on how to handle ordinal data in exogenous (predictor) covariates in pymc. Public Code, Open Source License

• PYMC Examples: GLM-missing-values-in-covariates Full Bayesian workflow example on how to handle and impute missing data in exogenous (predictor) covariates in pymc. Public Code, Open Source License

These are the result of a deep-dive investigation into a particular dataset, either to learn about that data or to fully demonstrate a suite of methods to tackle a new challenge.

• Oreum Case Study: Lung Survival regression of a lung cancer mortality study with conventional full-period observations. Uses a novel Accelerated Failure Time architecture that handles right-censoring, missing data, ordinal covariates and a linear regression submodel. Public Overview Slides, Commercial License

• Oreum Case Study: ONS Survival regression of a multi-year England & Wales ONS aggregated deaths-only dataset. This is a very compromised dataset but interesting to tackle, and requires a highly novel modified AFT architecture to allow for right-truncation (not right-censoring). Public Overview Slides, Commercial License

• Oreum Wholelife Simulator Simulate the cash value of WholeLife Insurance Policies using a parameterised survival curve including financial activities (investments, drawdowns). Publicly Readable Code, Commercial License

These are supporting repos to aid high-quality software development on client projects

• Oreum Core Core tools used by Oreum Industries on client projects to make data science work faster, more repeatable and consistent etc. Includes various data diagnostics, EDA plots, model factories, Bayesian model evaluation etc. Public Code, Open Source License

• Oreum Template A reusable template project structure - all our projects adhere to this. Publicly Readable Code, Commercial License

Oreum Industries © 2024