Project Title: Cognitive Hurdles In Note Management For Dyslexic Learners

​

Theme Chosen: AI For Sustainable Living

Team Name: Zero and One

Details of Team: Hari Balaji H

Rex Clement D

Muthuram S

​

​

​

Problem Statement & Need Analysis :

Several digital apps (like jamworks , Notion, and Evernote) exist, but they are not specifically designed to accommodate the needs of dyslexic learners.

�They do not include:

• OCR (Optical Character Recognition): To extract text automatically from handwritten or printed documents.
• Document Summarization: To generate concise, easy-to-read summaries of large content.
• Note Organization Intelligence: To categorize notes based on subject and chapterwise
• (RAG)-generative model to produce accurate, context-aware answers to user questions (interactive Q&A over the student’s notes).

​

Existing Solution and Its Limitation :

Individuals with dyslexia particularly struggle due to issues in reading fluency, spelling, and information retention, making traditional note-taking and studying methods less effective and individuals with dyslexia — face significant challenges in maintaining their notes in an organized and structured way.

AI-Driven Solution :

OCR for Dyslexia Support:�Converts handwritten or printed text into editable digital notes. Fine-tuned OCR handles varied handwriting, helping dyslexic users interpret notes accurately.

​

AI Summarization:�Creates concise summaries of long texts (notes, papers,) for quick understanding, text-to-speech allowing users to listen notes in their preferred language (Tamil & English).

​

RAG-Based Q&A:�Retrieval-Augmented Generation enables context-aware answers from stored notes using embeddings for accurate, interactive learning(Q&A).

​

Smart Note Organization:�categorizes notes by subject and chapter, with integrated checklists to manage study progress.

Technical Approach:

• Use a NoSQL database (like MongoDB) for flexibility,store as a document with metadata for easy organization.

​

• Handwritten OCR uses image preprocessing and a CRNN-CTC model to recognize handwritten text.

​

• A Python tool processes a file and uses Google's Gemini API to generate Tamil and English summaries optimized for dyslexic users.

​

• RAG (Retrieval-Augmented Generation) is an AI method that retrieves real data and then generates accurate, context-based answers using it.

Data & Model Training

Implementation

1. Developed a RAG-based Question Answering System using Python with FAISS retrieval and a LLaMA generator for context-aware, research-focused responses.

​

2. Fine-tuned the meta-lambda-3-1-8B-Instruct:novita embedding model on a custom Q&A dataset to improve semantic relevance and retrieval accuracy.

• Integrated a RAG system using FAISS and LLaMA to generate context-aware answers from notes and external sources.

​

• Recognized text and associated media are stored as documents in a NoSQL database (MongoDB) with metadata (class, subject, chapter).

​

• Handwritten text is extracted using Handwritten Optical Character Recognition (OCR) with a CRNN-CTC model.

​

• Users upload handwritten notes or capture images via camera.
• summarizer ,generates a clear, bilingual (Tamil and English) summary of the input file using the Gemini model with clean formatting and student-friendly language.

Future Scope & Scalability

1. Sora by OpenAI is an AI model that turns text descriptions into realistic, high-quality videos showing motion and detail.

​

2. It uses advanced transformer and diffusion techniques to generate smooth, consistent video frames that match the prompt.

​

Resource requirements:

• Python
• Hugging face model (embedding model)

• RAG concept
• Gemini model (2.5 flash)
• Sora (open ai)

RAG Architecture

Feasibility

1. Hallucination: The RAG model may retrieve inaccurate or outdated information, leading to misleading or incorrect answers.

​

2. Retrieval Quality Issues: Poorly fine-tuned embeddings or weak similarity searches can result in irrelevant document retrieval, reducing answer accuracy.

​

The AI-driven system effectively enhances learning accessibility and efficiency by combining OCR, summarization, and RAG-based Q&A capabilities. It supports dyslexic users by converting handwritten notes into readable digital text, provides concise summaries and multilingual text-to-speech for better comprehension, and enables intelligent, context-aware question answering. Additionally, smart note organization streamlines study management, making the overall solution a personalized, inclusive, and interactive learning support.

References:

Conclusion:

Links:

https://www.americanbrainfoundation.org/diseases/dyslexia/

https://en.wikipedia.org/wiki/Dyslexia