MedPath

MedPath Inspiration: Every day, patients leave hospitals with instructions they don’t fully understand. Nearly half struggle to interpret discharge instructions, which can lead to confusion, missed care steps, and preventable complications. Through conversations with patients, we realized the issue wasn’t access to healthcare—it was understanding it. In New Jersey, there are around 2.2 million immigrants and everyone might not understand or know English.

That led us to a simple question: What if every patient could leave care knowing exactly what to do, what to watch for, and when to act?

What We Built: MedPath is an AI-powered discharge clarity system that transforms complex medical instructions into simple, structured care plans.

Users paste instructions, and MedPath generates: Step-by-step actions in plain language Explanations of why each step matters Risks of not following instructions Warning signs and when to seek urgent care Multilingual support MedPath turns overwhelming clinical language into clear, actionable guidance.

How We Built It: MedPath uses the Anthropic Claude API to transform unstructured medical text into structured care plans. Key components include prompt engineering for clarity and safety, a web-based interface for input and output, and a translation layer for multilingual accessibility. A modular design organizes outputs into care steps and emergency guidance. The system goes beyond summarization—it restructures medical instructions into actionable decision support.

We model the transformation as: (instructions) → {steps, risks, warnings}

Challenges We Faced: Early versions only simplified text, so we had to redesign prompts to detect risks and provide meaningful guidance. Ensuring safety and clarity was critical—outputs needed to be calm, non-alarming, and patient-friendly. We also shifted from information delivery to structured, step-based action planning. Finally, we built the MVP in under 24 hours, prioritizing core functionality over UI polish.

What We Learned: We learned that understanding user needs matters more than adding features. AI is only useful when outputs are structured and actionable. In healthcare, clarity and safety matter more than complexity. Most importantly, the hardest part isn’t generating information—it’s making it usable in real life.

Built With: JavaScript / HTML / CSS Anthropic Claude API Prompt engineering pipeline Lightweight web interface Translation system