Real time biomedical monitoring applications, such as epileptic seizure or cardiac arrest detection, require low-latency sensing and processing at ultra-low-power budgets. In this overview paper, we review two main categories of biomedical systems: open-loop continuous real time monitoring for offline processing and closed-loop real time actuation for rehabilitation. We discuss their requirements and describe existing approaches to improve their system-level performance. We also compare and provide examples of various data compression techniques for open-loop systems. We then discuss several applications with embedded processing and explore how co-design of the full system (sensor, readout circuits and processing circuits) can yield significant benefits, making the case that such systems should be designed and optimized end-to-end.