perceptron.ai - empowering accessibility, available universally.
Inspiration
- Shocked by the lack of accessible tools in education despite legal mandates
- Personal encounters with students struggling due to inaccessible content
- Power of AI to democratize education for all learners
What it does
- Adapts content dynamically: Converts PDFs/images to colour-blind-friendly formats
- Audio inclusivity: Transforms text (PDFs, JPGs) into natural-sounding WAV for visually impaired users
How we built it
- AI backbone
- Deep learning-based encoder-decoder CNN model built with PyTorch
- One-hot encoding of color blindness type (protanopia, deuteranopia, tritanopia) used as conditional input
- Supervised learning with image-to-image translation approach
- Deep learning-based encoder-decoder CNN model built with PyTorch
- Color Correction
- Data-driven correction using paired original and corrected images
- Correction learned end-to-end from training data, not rule-based or traditional algorithms
- Data-driven correction using paired original and corrected images
- Infrastructure
- Frontend: React + TypeScript
- Backend: FastAPI
- Frontend: React + TypeScript
Challenges we ran into
- The Human Feedback Gap: Tight timelines made it hard to test with real users during development (e.g., no access to students with disabilities in the hackathon window).
- Edge Cases in the Wild: Real-world PDFs broke our pipelines (e.g., handwritten notes, complex tables, or stylized fonts).
Accomplishments that we're proud of
- Built a Functional Prototype in Record Time: Developed an end-to-end solution in just less than 2 days.
- Potential for extensive real-world use: Our solution is unique and is free to use.
What we learned
- AI can't Replace Human Judgement: Our initial AI-powered system made error on complex diagrams. We realised that human judgement is still needed for certain decision making.
- Accessibility is Multi-Dimensional: Simple color correction isn't enough - we needed dynamic solutions for dyslexia, low vision, and motor impairments.
What's next for perceptron.ai
Next 6 Months
- Launch institutional pilot programs targeting schools and universities with high populations of color-blind and visually impaired students
- Establish partnerships with special education departments to integrate our solution into existing accessibility workflows
Next 12 Months
- Scale our user base to serve 1,000+ students and educators worldwide
- Expand platform capabilities based on pilot feedback to support additional learning disabilities
Built With
- fastapi
- git
- javascript
- numpy
- pandas
- pdf2image
- pymupdf
- python
- pytorch
- pyttsx3
- react
- scikit-learn
- typescript
Log in or sign up for Devpost to join the conversation.