- Role
- Personal project
- Year
- 2026
- Status
- Live
- Stack
- Cloudflare Workers AI · Llama 3.3 70B · Astro · Cloudflare D1 · Turnstile · JavaScript · Vite
- body zones to choose from
- 14
- poses per suggestion
- 3
Overview
Describe your pain area and Yoga-Doc suggests yoga poses to ease muscles and reduce pain. Tap a zone on the body map and get three AI-suggested, beginner-friendly poses, each with instructions and a breathing note.
The project is open-sourced on GitHub, and a live version runs on this site.
Approach
- Fixed choices, not free text. The visitor picks one of 14 body zones. The browser sends only a zone id; the prompt is built on the server, so visitors cannot inject instructions.
- Safety-minded by default. The model is instructed to suggest only beginner-friendly poses and to remind the user to stop if they feel sharp pain. Every result carries a “not medical advice” note.
- Structured output. Each suggestion follows one format: pose name, how to do it, and a breathing note.
- Graceful fallback. If the live AI is unavailable, rate limited, or the daily quota is used up, the app shows sample poses and says so.
Architecture
The open-source version is a JavaScript app built with Vite. The live demo on this site runs the same interface against Cloudflare Workers AI.
- Interface: a static app served from this site.
- API: a server-side route that maps the zone id to a prompt and streams the answer back as plain text.
- Model: Llama 3.3 70B on Cloudflare Workers AI. No third-party API key is involved.
- Protection: origin check, per-IP rate limit (10 requests a minute), Cloudflare Turnstile human check with a short-lived signed session, and a per-day request cap with a kill switch stored in Cloudflare D1.
- Privacy: the request log stores no prompts and no raw IP addresses, only a salted hash.
Demo
Try the live demo. It asks for a quick human check before the AI runs.