AccessiCore

The AI-Powered Semantic Web Repairer

AccessiCore is an open-source browser extension that bridges the "Accessibility Gap." By using a Local Small Language Model (SLM), it re-indexes poorly designed, non-compliant websites in real-time, providing a conversational interface and automatic ARIA-label "repairs" for the visually impaired.

The Problem

Despite the Web Content Accessibility Guidelines (WCAG), millions of essential websites—government portals, legacy forums, and small business sites—remain inaccessible.

• Broken Layouts: Standard screen readers get lost in "Div-soup."

• Missing Context: Images without alt tags and buttons without ARIA labels make navigation impossible.

• Information Overload: Users have to listen to 10 minutes of headers just to find one "Submit" button.

The Solution: AccessiCore

AccessiCore doesn't just read the code; it understands the intent. By running a quantized SLM directly in the browser via WebGPU, it creates a "Semantic Map" of the page, allowing users to interact with any website through a natural language, voice-first interface.

Key Features

1. Semantic DOM Re-Indexing

The extension parses the active DOM and converts complex structures into a simplified, hierarchical JSON map that the AI can interpret instantly.

2. "Talk to the Web" Interface

Users can press a hotkey and ask questions like:

• "Where is the registration form?"

• "What are the eligibility requirements mentioned on this page?"

• "Submit this form using my saved profile."

3. Live ARIA Injection

The local vision-capable model identifies icons and unlabeled buttons, automatically injecting proper aria-label and alt attributes into the live DOM before the screen reader reaches them.

4. 100% Privacy & Offline-First

No data ever leaves the user's machine. All processing is done locally using Transformers.js and WebLLM, ensuring that sensitive data on government or banking portals remains private.