We Built an AI Cardiologist Into a Patch

What if your ECG could just tell you what's wrong? We built a cardio chatbot that pairs STMicroelectronics biosensors with generative AI to turn 10-second signals into plain-language insights — fast enough for real clinical use. The secret isn't throwing raw waveforms at a large model. It's context engineering: smart preprocessing, clinically grounded features, and prompts that force the LLM to reason with P waves, QRS width, QT intervals, and heart rate signatures. We break down the full pipeline — skin patch capture, NeuroKit2 extraction, structured GPT-4 prompts, and a conversational loop that clarifies findings on the fly. Perfect F1 on ventricular tachycardia and AV block. Strong results on AFib and normal sinus rhythm. And honest gaps on electrolyte-driven states like hyperkalemia — because that's what real-world performance looks like. If you care about wearable ECGs and AI-powered healthcare, this one's for you.