Silver

Phone-sized resurrection of camera + voice for older adults — large buttons, honest demo mode, optional vision URL.

Comment

Inspiration

Smart glasses and clinic video magnifiers already tried to help people read what is in front of them. Glasses were set aside for many people; desk magnifiers help, but they stay on a desk. Silver brings that same job back onto a phone someone already owns: a camera, a voice, and a simple face.

What it does

  • Mobile page with large labeled buttons, large type, light high-contrast layout.
  • Type and hear — browser speech for typed text (no camera, no API key).
  • Open camera or choose a photo — image stays on the phone unless vision is configured.
  • Read text / Describe the scene:
    • No VISION_API_URL: honest demo sentence that Silver did not read or describe the photo.
    • With VISION_API_URL: the Cloudflare Worker POSTs the photo; the page speaks the returned text.
    • Vision call failure shows an error message, not fake demo text.

How Kiro was used

The repo keeps .kiro/ on purpose (not gitignored):

  • .kiro/steering/ — product, stack, layout, honesty, accessibility, vision contract
  • .kiro/specs/large-button-ui/ — requirements, design, and tasks for the page
  • .kiro/specs/read-aloud/ — speech and photo reading
  • .kiro/specs/scene-description/ — scene description path
  • .kiro/hooks/ — PostFileSave typecheck, honesty reminder on copy changes, and a Stop hook that runs npm test

Specs are requirements-first (user stories and EARS-style WHEN / THE SYSTEM SHALL). Checked tasks are built; unchecked items (on-device OCR, Bluetooth glasses, saved voice speed) are not claimed as done.

Challenges

  • Staying honest under demo pressure: no pretending the app saw a photo when no vision service is wired.
  • Separating "configured" (VISION_API_URL set) from "worked" (the service actually returned text).
  • Phone speech quirks: some devices only speak from the last tapped button; the UI includes Hear it again, Stop, and speed controls.
  • Shipping as one Cloudflare Worker plus static assets so the page and API deploy together.

Accomplishments

  • Public MIT repo with steering, specs, hooks, tests, and a live workers.dev demo.
  • Honest demo mode and fail-closed vision errors.
  • Resurrection framing that is product-real, not costume UI.

What we learned

Requirements-first specs plus an honesty steering file keep the agent from inventing features. A vision stub that admits it does not look at the photo is more useful for a hackathon demo than a fake OCR claim.

What's next

Optional real vision endpoint behind VISION_API_URL. On-device OCR and Bluetooth glasses stay unchecked tasks and are not claimed today.

Limits

The public deploy currently runs without VISION_API_URL (demo mode). It will not OCR or describe photos. Type and hear works in a supporting browser. Silver is not a medical device. There is no Bluetooth glasses pairing and no built-in OCR.

Built With

Share this project:

Updates

Submission history