NeuroRead

Empowering neurodiverse readers with clear, simplified, and private article comprehension powered by Gemini Nano.

Comment

Inspiration

As someone passionate about accessibility and AI, I wanted to create a tool that helps neurodiverse readers, people with ADHD, dyslexia, or cognitive overload, consume online content more comfortably. While existing reader modes focus on visual formatting, few address cognitive accessibility, such as simplifying complex text or summarizing dense material. With the release of Gemini Nano, Chrome’s on-device AI model, I saw an opportunity to build something that works locally, privately, and efficiently without relying on external servers.

What it does

NeuroRead is a Chrome extension that uses Gemini Nano’s built-in AI APIs to:

  1. Summarize long articles into short, easy-to-digest insights

  2. Simplify text at different reading levels for better comprehension

  3. Run entirely on-device, preserving privacy and working even offline

  4. Fallback gracefully to a local summarizer if Gemini Nano isn’t available

How I built it

  1. Designed the user interface with HTML/CSS for simplicity and focus.

  2. Used JavaScript (ES6) with Chrome Manifest v3 to build the extension.

  3. Integrated Chrome’s Built-in AI APIs (Summarizer and Rewriter) for summarization and simplification.

  4. Added logic to detect whether Gemini Nano was available and fall back to a local summarizer.

  5. Tested and refined the summarization quality using multiple article types: news, blog posts, and research summaries.

Challenges I ran into

1, API Availability: Many Gemini Nano APIs are still experimental; testing required Chrome Canary builds and enabling multiple flags.

  1. Fallback Logic: Ensuring smooth transitions between local and on-device AI summarization.

  2. Text Extraction: Filtering main content while skipping headers, ads, and quotes required careful DOM selection.

  3. Model Feedback: Without API telemetry, debugging model behavior had to rely on qualitative output testing.

Accomplishments that I'm proud of

  1. Built a fully functional Chrome extension that runs without external servers.

  2. Created a privacy-first AI summarization pipeline.

  3. Designed a user experience tailored for neurodiverse users and general readers alike.

  4. Learned how to leverage emerging web AI features to improve real-world accessibility.

What I learned

  1. How to work with experimental browser APIs and the Gemini Nano runtime environment.

  2. The difference between client-side and hybrid AI approaches in Chrome.

  3. The importance of UX design for accessibility, ensuring minimal cognitive load.

  4. How to use content filtering to target article text while ignoring headlines and navigation links.

  5. How on-device AI can transform accessibility, no internet dependency, no privacy trade-offs.

What's next for NeuroRead

  1. Multimodal Input: Integrate image and audio understanding using the multimodal Prompt API to help summarize or describe visual content

  2. Hybrid AI Mode: Combine Gemini Nano’s local model with the Gemini Developer API for enhanced context depth when online

  3. User Profiles: Allow readers to set preferences for tone, reading level, and simplification type (e.g., “academic,” “casual”)

  4. Highlight + Summarize: Enable users to select text directly on a page and generate focused summaries

  5. Chrome Web Store Launch: Package and publish the extension publicly once Gemini Nano APIs are officially released

Built With

Share this project:

Updates