I need an AI-driven “do / don’t” engine that screens user-generated text and flags or blocks anything that looks like spam. The scope is strictly text-based content moderation—no images or videos for now—and the only category I care about at this stage is spam messages. Your job is to design, train, and deploy a model (or rules-plus-model hybrid) that can make real-time accept / reject decisions with a confidence score I can log. An easy-to-consume REST or GraphQL endpoint is ideal so I can plug it into my current backend. Please include a short README that explains the input format, expected response, and any threshold settings I can tweak. I’ll supply sample data to get you started, but I’m open to suggestions on public corpora or augmentation techniques if you think they’ll improve accuracy. What matters most is low false-positive rates, fast inference, and clear instructions so I can maintain or retrain the system later.