EchoClass

Inspiration

Many international and neurodiverse students struggle with traditional note-taking tools that assume everyone learns the same way. Having ESL students in our team, we noticed that class notes often contain repeated misunderstandings, unexplained terms, or inaccessible formats. We wanted to create a tool that makes learning inclusive, adaptive, and fair, giving every student an equal chance to understand and review their notes.

What it does

EchoClass is an AI-powered note-taking assistant that enhances the learning process from comprehension to review. It automatically: • Detects and highlights common mistakes in student notes • Explains complex or technical terms in plain English or bilingual mode • Converts typed math expressions into LaTeX for clarity • Generates Anki/Quizlet-style flashcards for spaced repetition • Supports dyslexia-friendly text-to-speech function to make reading easier Together, these features reflect the complete learning loop: Correct → Understand → Document → Test.

How we built it

We built EchoClass using: • FastAPI + Python for the backend server • Gemini 2.5 Flash for real-time text analysis, translation, and LaTeX generation • JSON Schema prompting to structure Gemini’s output into mistake reports, explanations, and flashcards • HTML + JavaScript frontend for live interaction with the model • .env + dotenv for secure key management Each module (mistake detection, translation, LaTeX, and flashcards) connects through a unified API layer, so teachers and students can access all features in one interactive web app.

Challenges we ran into

• Structuring Gemini’s responses into valid, machine-readable JSON while keeping natural language quality
• Synchronizing multiple features (mistake detection, translation, LaTeX) in one FastAPI endpoint
• Managing rate limits and environment conflicts between local Python versions and virtual environments
• Designing the UI to stay simple but still communicate analytical feedback clearly

Accomplishments that we're proud of

• Built a fully functional demo that detects errors, translates notes, and generates flashcards in real time
• Successfully integrated Gemini 2.5 Flash with FastAPI, making the backend scalable and modular
• Designed the system around inclusive education principles, ensuring accessibility for ESL and dyslexic learners
• Collaborated effectively as a team, combining AI engineering, UX thinking, and pedagogical insight

What we learned

We learned how to connect Gemini models with structured JSON outputs through FastAPI, improving our skills in prompt engineering and LLM integration. We also discovered how small prompt changes can greatly affect model clarity and consistency. On the social side, we realized that designing for ESL and dyslexic learners requires empathy and that accessibility is not an add-on but instead a responsibility. Building EchoClass reminded us that AI can empower every learner when it’s built with both technical precision and human understanding.

What's next for EchoClass

We aim to make EchoClass a fully integrative learning ecosystem. Our next steps include: 1. Handwritten input integration: enable scanning of paper notes with CNN-based handwriting recognition, and connect EchoClass with existing iPad note-taking apps for direct input. 2. Bias detection and analysis: detect bias or unbalanced representations in note content and AI outputs. 3. Enhanced flashcard generation: improve Anki/Quizlet export with adaptive review frequency based on student learning patterns. 4. Class-level analytics: allow teachers to track collective misunderstanding trends, identify topics needing reinforcement, and auto-generate quizzes or lesson improvements.

Ultimately, EchoClass will evolve from a note assistant into a learning intelligence platform that empowers both students and teachers to create more inclusive and personalized education.