Designing for life or death: building data products for biopharma operations with Fathom

We’ve gotten used to algorithms making decisions in our daily lives. Yet, statistics and its modern incarnation — artificial intelligence and machine learning — are all inherently probabilistic. Probabilistic processes have the potential for error.  Unhandled edge cases and a base error rate are normal parts of any AI solution but present a unique risk when the intended use case is medicine. In this lightning talk, I will discuss the challenge of designing and deploying AI/ML to support biomanufacturing operations, outlining some of the algorithmic constraints and architecture choices Fathom has employed to manage risk while improving the manufacture and distribution of life-saving drugs.