About

Built in a day. Still being built.

HealthDojo started as a one-day prototype at a healthcare AI hackathon and is now an open, early test ground for home-health AI.

01 · Origin

A hackathon finalist

HealthDojo was built solo in one day at the Healthcare AI Hackathon hosted by Pear VC with NEA and Cathay on September 26, 2026, where it was selected as a finalist.

The question it started with: how do you test a home-health AI before anyone relies on it in a patient's home? The answer was to turn published clinical guidelines into synthetic homes with exact labels, then see what vision models catch and miss.

Since the hackathon, the site has been rebuilt and the photo benchmark has grown to more than 20 vision models across two benchmarks.

02 · Builder

Pranay Madan

Builder of HealthDojo. Open to conversations with clinicians, home-health teams, and AI companies who want to pressure-test models before real-home studies.

Emailmadan.pranay@gmail.com LinkedInlinkedin.com/in/pranaymadan Codegithub.com/pistachiopranay/healthdojo

03 · Stack

What it runs on

Model access
OpenRouter
25+ vision models from Anthropic, OpenAI, Google, Meta, Qwen, Mistral and more, ranked on one leaderboard.
World model
World Labs Marble
Turns each synthetic home into a walkable 3D world with collision meshes, right in the browser.
Image generation
OpenAI, Gemini and Stability AI
Clean rooms first, then hazards painted in one label at a time.
Voice
ElevenLabs
Narrates each walkthrough step.
Deployment
Cloudflare
Served from Cloudflare; the code and benchmarks are open on GitHub.

HealthDojo is a prototype. All data is synthetic, there is no patient data, and it is not clinically validated. Clinician review and real-home studies remain necessary before clinical claims. See Limits.