AIdeas: ReadEase AI — A Dyslexia-Friendly Browser Extension Powered by Amazon Nova
Dyslexia affects 1 in 10 people, yet almost no web content is designed with them in mind. This is how I built a Chrome extension and serverless AWS backend that instantly reformats any webpage for readability and uses Amazon Nova family of models on Bedrock to simplify complex text, explain difficult passages, and interpret visual content - all on demand, privately, and within the AWS Free Tier.
App category : Social Impact
Team Members : @anas13, @blight
My Vision
Reading online should not feel like a battle.
For people with dyslexia, a typical webpage can be overwhelming — dense paragraphs, small tight-spaced fonts, low contrast, cluttered layouts, and complex vocabulary combine into a wall of cognitive friction. Every article, form, news story, and product page becomes harder than it needs to be. ReadEase AI is a Chrome browser extension and lightweight web dashboard designed to act as a personal AI reading companion. It is built around four core capabilities:
- Reformatting any webpage into a dyslexia-friendly layout using local CSS controls — no AI call needed, no latency, no cost.
- Simplifying selected text using Amazon Nova Lite through Amazon Bedrock, rewriting complex sentences into shorter, clearer language while preserving meaning.
- Explaining difficult passages on demand — a user can highlight a paragraph and request a calm, step-by-step plain-language explanation.
- Interpreting images, charts, and diagrams in simple language, so visual content is no longer a dead end for readers who need more context.
The goal is a browser that adapts to the reader instead of forcing the reader to adapt to the page.
Why This Matters
Dyslexia is not a measure of intelligence. It is a difference in how the brain processes written language, and it affects an estimated 700 million people globally. Many of them develop workarounds — reading slowly, asking others for help, avoiding certain tasks entirely, or simply not engaging with content that matters to them.
The internet was supposed to democratize access to information. In practice, most of the web is designed for people who read quickly and comfortably. That leaves a large population underserved by default.
ReadEase AI addresses this in a way that is:
- Immediate — formatting changes apply locally in milliseconds, with no server round-trip.
- Non-intrusive — the original page is preserved; users can toggle ReadEase on or off at any time.
- Private — AI is only invoked when a user explicitly asks for help with a specific selection. No full-page scanning, no passive data collection.
- Low-cost to operate — the architecture is designed around the AWS Free Tier so that cost is never a reason to limit access.
This matters because accessibility tools should not be a luxury or a workaround. They should be built into the environments where people spend their time. ReadEase AI is a step toward making every webpage as readable as it can be for every person who visits it.
How I Built This
Architecture at a glance
ReadEase AI is a monorepo containing three parts: a Chrome extension, a Python backend, and a lightweight web dashboard.
Extension
The extension is built with React and Vite targeting Chrome Manifest V3. It is split into four layers:
- Popup UI — the main interface for toggling formatting options and triggering AI actions.
- Options page — persistent settings management, synced to local browser storage.
- Content script — injects CSS transformations into the active page. Uses a
TreeWalkerto find readable text nodes safely, skipping inputs, scripts, code blocks, and editable regions. AMutationObserverwith throttling handles dynamically loaded content. - Background service worker — mediates all API calls, manages the right-click context menu, and runs a local cache keyed by content hash and action type with a 24-hour TTL.
Formatting is handled entirely in CSS first: font family (OpenDyslexic), font size, line height, letter spacing, word spacing, color overlay, and focus mode. AI is never called for formatting — only for simplification, explanation, and visual interpretation.
Backend
The backend is a FastAPI application deployed to AWS Lambda using the Mangum adapter and exposed through Amazon API Gateway. Routes include:
| Route | Purpose |
|---|---|
POST /simplify | Rewrite selected text in simpler language |
POST /explain | Explain a passage with a chosen mode (simple, analogy, step-by-step) |
POST /summarize | Summarize a passage as bullets or a short paragraph |
POST /visual-explain | Interpret an image, chart, or diagram |
GET /health | Service status and configured model IDs |
POST /sessions | Log lightweight anonymized session events |
Every route validates input strictly — empty text, oversized payloads, and unsupported modes are rejected before any Bedrock call is made. Character limits per call (2,500–4,000 for simplify/explain, 5,000 for summarize) keep costs predictable.
Demo
What I Learned
1. Local-first is the right default
The most impactful thing ReadEase AI does — reformatting the page — costs nothing and takes milliseconds. Starting with CSS transforms before reaching for AI forced better decisions throughout the build. AI should amplify what a tool can do, not substitute for things that can be solved more simply.
2. Prompt design is product design
The quality of the simplification and explanation outputs is almost entirely determined by the system prompt. Getting the tone right — calm, respectful, not condescending — required iteration. The instruction "avoid childish tone" made a measurable difference in output quality. Prompt engineering is not a finishing step; it is core to what the product feels like.
3. Content scripts are harder than they look
Injecting CSS and DOM modifications into arbitrary third-party pages without breaking layouts is genuinely difficult. Sites with aggressive CSS specificity, shadow DOM components, and dynamic content loaders required careful defensive coding. The rule to prefer CSS injection over DOM replacement whenever possible was the right call — it eliminated an entire class of bugs.
4. Cost discipline shapes architecture
Designing around the AWS Free Tier was not a constraint — it was a design principle that made the architecture better. On-demand AI, hard character limits, aggressive local caching, and tiny session logs are all patterns that improve reliability and predictability, not just cost. Building cost-aware systems early produces better systems.
5. Accessibility is an engineering problem, not just a design problem
Most accessibility guidance focuses on visual design. ReadEase AI revealed that accessibility also lives in font rendering, text processing logic, content parsing, error handling (a crash during reading is especially disorienting), and the latency of AI responses. Every engineering decision had an accessibility consequence.
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article