Although the serverless environment has been gaining popularity over the past few years, its use for deployment of ML models has been largely unexplored. In this talk I'll try to change that and discuss our journey of using serverless (specifically AWS Lambda) to power Slido's NLP-enabled features by going through it all: the good, the bad and the ugly. Despite its limits, I'll try to show a couple examples where it enabled projects that would not make it to production otherwise.