Accelerate safely with machine learning model governance.

Machine learning is driving innovation in banking. But while it delivers a strong competitive edge, integrating machine learning with traditional model development processes raises some new challenges for compliance.

Dotscience brings machine learning into a robust model governance framework in order to help banks reduce their model risk.

Our solution makes machine learning models compliant and auditable, so that financial institutions can safely rely on them.

Designed for model development in Python and R, in Jupyter or RStudio environments

Dotscience provides a suite of tools for model governance. Our tools enable compliance with regulators and auditors; provide analytics on model behaviour; and improve robustness of modern model development processes.


Provenance Graph auto-generates a complete record of the model’s history in development. It includes all changes made to the model’s code and to any data ingested. Dotscience’s filesystem technology allows reference to snapshots of datasets, so that as training and validation data is cleaned, updated and engineered over time, the exact version used by the model remains both accessible and persistently coupled with the model code.

Provenance Graph forms a record for regulatory compliance, and an audit-ready paper trail.

Model explainability

Quants and developers can learn from every training run made by anyone in their team, with Dotscience Dashboard. Model metadata, such as parameter settings and metric scores, is captured per run and made available in aggregate. Interactive visualisations of this data provide insights into model behaviour and allow better decisions to be made in the optimisation process.

Dashboard shows interactions between parameters and output metrics: it can be used to isolate the effect of a single variable on model performance, as in attribution analysis, and to identify broader trends during optimization. By allowing developers to better understand their models, Dashboard protects against the risk of a model generating incorrect or misused outcomes.

Version control for data and code

Dotscience builds robustness and course-correction into the model development process with GitHub-like version control for both datasets and code. Unlike GitHub, Dotscience can version arbitrarily large datasets of the kind needed in financial modeling. Version control brings proven collaboration workflows and productivity gains to the model development process.

Our version control system also captures rich metadata — including model parameterization, metrics and summary stats — from every run, along with a commit of the code executed and data consumed. This full picture of each run enables reproducibility of any experiment, so that managers and collaborators can quickly identify, inspect and reproduce runs of interest.