Wouldn't a personalized RSS algo be a great use case for a simple LLM?