Keiro

Inspiration

I moved from Australia to Texas 18 months ago. I never dealt with a language barrier in a hospital, but I watched people who did. My family has hospital connections, and the stories were always the same: a patient describes their pain in broken English, translation gets lost, and the treatment plan misses the important information. Millions of people walk into hospitals every year and can't understand their own doctor. In an emergency, it's the difference between the right diagnosis and the wrong one.

What it does

Keiro is an AI-native healthcare assistant that speaks over 40 languages. A patient opens the chat and speaks to Kai in their native language, and Kai asks questions such as what's wrong, where it hurts, how severe it is, and how long it has lasted. After that, Keiro creates a clinical-grade report that is organized just like how a doctor would read a chart, including the patient's summary, chief complaint, symptom severity, and onset. The patient hands it over for their doctor to read and understand.

How we built it

Keiro runs on Next.js and Tailwind for the frontend, with Supabase managing authentication and patient data storage behind row-level security. Claude runs the conversational intelligence, Google OAuth handles sign-in, and we run Sentry/PostHog underneath everything to catch issues before they reach the patients. In order to ensure that a doctor receives a report that is actually useful rather than a translated transcript, we designed the intake format around how a real nurse evaluates a patient, starting with open-ended questions and getting to focus on severity.

Challenges we ran into

Getting the tone right took longer than the code. An AI assistant talking to someone in pain, in a language they're more comfortable in than English, has to feel calm and proficient at the same time. We rewrote Kai's conversation flow multiple times to land on something that felt like a person. We also had to get the security right before anything else. Patient data isn't something you patch later, so row-level security and OAuth were in place before we built anything else.

Accomplishments that we're proud of

Keiro is live and deployed at keiro.space. It generates quality clinical reports from real conversations, in multiple languages, end to end.

What we learned

Building for healthcare means building for someone at their most vulnerable. Every design decision has importance when the person on the other end is scared and can't fully explain what's wrong.

What's next for Keiro

We want to expand language coverage further, add live consult mode so doctors and patients can communicate in real time during an actual appointment, and start piloting with clinics that see high amounts of non-English-speaking patients.

Built With

• anthropic-api
• claude
• google-oauth
• next.js
• postgresql
• posthog
• react
• sentry
• supabase
• tailwind-css
• typescript
• vercel