Inspiration
Digital accessibility is often treated as a "nice-to-have" or a last-minute legal check, rather than a fundamental human right. We wanted to build a tool that bridges the gap between design and inclusivity, making it effortless for creators to ensure their content is readable for the 1 in 4 adults living with a disability while handling data responsibly by processing images in memory rather than saving them to a permanent database. This protects user privacy.
What it does
AccessCanvas is an AI-powered accessibility auditor. Users upload an image, and the app instantly performs two critical tasks:
Contrast Auditing: It uses computer vision to detect text and calculate contrast ratios against WCAG 2.2 standards, flagging "Critical" or "Warning" regions.
AI Alt-Text: It leverages Gemini 2.0 Flash to synthesize OCR data into natural, semantic alt-text and long descriptions for screen readers.
How we built it
Frontend: A responsive React (Vite) interface designed for speed and clarity.
Backend: A Python/Flask API that handles image processing via OpenCV and Pillow.
Infrastructure: The backend is fully containerized using Docker to manage system-level dependencies like Tesseract OCR.
AI Orchestration: Integrated OpenRouter to access state-of-the-art LLMs for metadata generation.
Deployment: Hosted on Render using a decoupled architecture (Static Site + Web Service).
Challenges we ran into
The biggest hurdle wasn't the code, it was the environment. Deploying Tesseract OCR required a deep dive into Docker and Linux library management to resolve versioning conflicts (like the libgl1-mesa-glx error on macOS). We also faced the CORS issues. Navigating cross origin headers between a decoupled frontend and backend required a robust, production-grade security configuration.
Accomplishments that we're proud of
We are incredibly proud of successfully building a production-ready Docker pipeline that works seamlessly in the cloud. We managed to turn a complex, multi-step accessibility audit into a "one-click" user experience that produces high-fidelity, human-readable AI results that doesn't replace the artists work but rather assists in it.
What we learned
We learned that 90% of development is deployment. Building a local app is one thing, but managing infrastructure, containerization, and server-side dependencies is where the real engineering happens. We also gained a deep appreciation for the technical nuances of inclusive design.
What's next for AccessCanvas
The goal is to scale AccessCanvas into a browser extension that audits live websites in real-time. We also want to implement bulk processing for social media managers and deeper integration with design tools like Figma to catch accessibility issues before they ever reach the web.