Polaris

Inspiration

Reading is something most people never think about, but for people with dyslexia every email, textbook and form takes extra effort. Most reading tools live in a single browser tab or app, while reading happens everywhere. We wanted a tool that meets you wherever the text is, adapts to how you read, and guides you from A to B without reading for you. That's where the name comes from: Polaris, the North Star.

What it does

Polaris is an AI reading companion for macOS, paired with a web app.

Select text in any app. Polaris opens a reformatted reading view with your preferred font, size, letter spacing, line spacing and background tint.
Read aloud with word-by-word highlighting. Each word lights up as it's spoken. Arrows let you step forward and back, and you can set the reading speed.
Simplify. Gemini rewrites the text to your chosen reading level (simple, standard or advanced).
Scan images and screenshots. Text inside images is extracted on the device and read the same way.
Reading assessment quiz. A short quiz on the web app suggests your best settings (font, spacing, voice and reading level) and saves them to your account. The Mac app then picks them up.
Choose a voice. Preview ElevenLabs voices on the web app and pick the one that sounds right to you.
Save to your account. A hotkey saves a selection, along with where it came from, to your web dashboard.

How we built it

Mac app (Swift): the macOS Accessibility API captures selected text and its context (app, window title, URL). Apple's Vision framework (VNRecognizeTextRequest) handles OCR on the device, so it's free, fast and works offline.
Pipeline: Scanner → Reader (OCR) → Simplifier (Gemini) → Text-to-speech.
AI: Gemini simplifies text and helps turn quiz answers into recommended settings. ElevenLabs provides natural-sounding voices.
Web app: a quiz and settings UI built around dyslexia-friendly design guidelines, with an SQL backend, OAuth, and a REST API (GET/PUT /api/settings, POST /api/save) secured with bearer tokens so the Mac app and website share one profile.
Privacy by design:
- Sensitive apps (password managers, banking, Messages) are blocked from capture.
- Emails, phone numbers and card numbers are redacted on the device before anything reaches the API.
- Data stays on the Mac by default, and syncing is opt-in.
- A visible "paused" indicator and hotkey stop capture.
- Screenshots show a confirmation before being sent to Gemini.

Challenges we ran into

Not just a wrapper: our first version was "select text, ask an LLM." We pushed past that by focusing on what a chatbot can't do: capturing context from any app, personalized settings, synced read-aloud, and on-device privacy.
Mixed laptops: only some of us had Macs, so we split the work into a Swift app and a web app that talk through a strict settings format agreed early on.
Keeping both sides in sync: the Mac app rejected a settings object if any field was missing, and macOS voice IDs mean nothing to a website. We moved voice selection to ElevenLabs voice IDs that work on both sides.
Syncing the highlight with the audio so the right word lights up at the right moment.
Security: an app that reads your screen raises fair privacy questions, so we made privacy a core feature instead of an afterthought.

Accomplishments that we're proud of

A working pipeline from any app on the Mac to a reformatted, read-aloud, simplified view
One profile shared across the Mac app and the web app
Design decisions based on dyslexia research, not guesswork
Privacy features built in from day one