ACCESS-AI

About the Project

Inspiration

Accessibility is often treated as an afterthought in software development. Yet over 1 billion people worldwide live with some form of disability that makes the web hard to use. We were inspired by this gap: what if AI could act as a universal accessibility layer, instantly making any website more usable without waiting for developers to redesign their apps? That’s how AccessAI was born — a browser extension that combines computer vision, natural language processing, and voice interaction to bridge the accessibility divide.

What We Learned

How to integrate Chrome Extension APIs (Manifest V3) with a backend service. Applying OCR and computer vision to detect images and missing alt-text. Using LLMs to summarize complex text into simpler, more digestible content. Implementing voice input/output with the Web Speech API for real-time navigation. Balancing hackathon constraints: deciding what features to prioritize for a working demo.

How We Built It

Frontend (Extension Layer): JavaScript/TypeScript to inject a content script into any page. UI popup built with simple React components for toggling features. Backend (AI Processing): FastAPI (Python) server exposing endpoints for /summarize_text and /describe_image. OCR via Tesseract + AI captioning models. Summarization via OpenAI GPT-5 API (with adjustable detail levels).

Voice Navigation: Browser’s Web Speech API to capture user voice commands. Text-to-speech for reading summaries and alt-text aloud. Command parsing to trigger actions (e.g., “click checkout”).

Data Flow: User Voice  →Speech to Text Extension  →REST API Backend AI  →Response Extension  →Speech User Voice Speech to Text

Built With

• ai
• ai-powered
• ai-processing)-frameworks-&-libraries:-react-(extension-popup-ui)-fastapi-(backend-rest-api)-tesseract-/-opencv-(ocr-+-computer-vision)-apis-&-ai-models:-openai-api-(summarization
• alt-text-generation
• assistant
• built-with-languages:-javascript-/-typescript-(chrome-extension
• coding
• collaboration)
• cursor
• fallback-for-local-ai)-vercel-/-heroku-(for-hosting-backend-demo)-tools:-github-(version-control
• frontend-logic)-python-(backend
• natural-language-understanding)-web-speech-api-(speech-to-text-and-text-to-speech-in-browser)-platforms-&-services:-chrome-extension-(manifest-v3)-hugging-face-models-(optional