2025-09-14 –, Ballroom 1
This is an overview of the tips and tricks I learned while bringing a mathematical optimization model from concept to prototype to production.
At its core, the math model is a multi-dimensional knapsack problem, with tens of thousands of items moving between warehouses with up to a hundred user-defined constraints. The model is solved with Google's OR-Tools and is deployed to AWS Lambda.
Some of the important ideas I want to share are how to build a test suite to ensure that a model of this type is internally consistent and how to handle errors-as-data for your end users. I also learned a really nice way to configure logging in python.
I learned some things about structuring my data science model. These things are kind of neat, and definitely worth sharing.
The test suite for an optimization model needs to ensure internal consistency of the model - so each constraint type is tested in two different ways, once for evaluation and once for optimization. If the individual constraints are consistent, then user requests with multiple constraints have fewer sources of failure. Also, when checking my test coverage reports I discovered the joys of branch testing.
To communicate errors as data back to end users, I created a simple InformativeError data model, and included it as an attribute on every response. So the dashboard application will always have a consistent way to display fatal and non-fatal errors back to the user.
Finally, when building a logging framework, I discovered how to output structured JSON logs, and I have a dictionary config that specifies different filters for logging from the CLI, from Jupyter and from lambda.
My background is in applied mathematics and data science. I have a PhD and 15 years industry experience. Currently working as a data scientist at a fintech startup where my interest is in building maintainable and robust models for deployment.