Inspiration
Nurses spend a large part of every shift answering call lights, and many of those calls are for small things like adjusting the bed, getting water, or asking when the next meal is. Each one pulls a nurse away from patients who might need urgent care. We want to build something that handles the small requests and makes sure the serious ones never get lost in the noise.
What it does
傷心小狗 is a planned AI bedside assistant for patients, paired with a dashboard for nurses and doctors.
For patients: They can ask simple questions (meal times, visiting hours, room controls) and get help with minor requests without pressing the call button. For care staff: The assistant will decide whether a request needs a human. Anything involving pain, breathing, falls, confusion, or medical concerns goes straight to a nurse, labeled by urgency. Low-priority requests are logged so nurses can batch them. Dashboard: It will show each patient’s status at a glance: who needs help now, who has a pending request, and a history of each interaction.
The assistant will not give medical advice. When it’s unsure, it escalates to a nurse.
How we built it
We’re at the design stage. The planned architecture:
A patient-facing chat/voice interface An LLM for conversation, plus a triage layer that classifies each request as AI-handleable or nurse-required A real-time nurse dashboard that receives alerts as they happen Challenges we ran into
So far the main challenge has been design rather than code:
Where to draw the line: Deciding which requests the AI can safely handle. We’ve chosen to over-escalate rather than risk missing something. Avoiding medical advice: Figuring out how to keep the assistant helpful without it diagnosing or recommending treatment. Accomplishments that we’re proud of Defining a safety-first escalation model that defaults to bringing in a human Scoping a concept that targets a real, specific pain point in nursing workflows
What we learned
Healthcare AI depends more on knowing when not to act than on being clever Much of a nurse’s workload comes from small, low-priority interruptions
What’s next for 傷心小狗
Build the prototype: patient interface, triage logic, and nurse dashboard Get feedback from nurses to set the escalation rules Add multilingual support and accessibility features Explore integration with hospital call-light and EHR systems
