clearmind pro

Inspiration

In India, 250M+ students prepare for brutal exams by re-reading dense textbook paragraphs with near-zero comprehension — memorizing keywords instead of building mental models. When they get stuck, chatbots dump a wall of text without ever diagnosing where the misunderstanding happened. We wanted to build the tutor we wished we'd had: one that asks questions instead of spoon-feeding answers.

What it does

ClearMind Pro is an AI master tutor. Its tutor Luna calibrates to your academic tier (Class 6 → B.Tech, MBBS, self-directed), auto-generates your syllabus, and guides you through an active mastery cycle: a Socratic voice coach that asks guiding questions, a smart whiteboard where you draw handwritten steps and AI marks exactly where your reasoning broke (red/green pens), on-the-fly 60 FPS physics/chemistry labs, a "Feynman arena" where you prove mastery by teaching Leo — a curious AI kid who asks trap questions — plus 3D flashcards with Anki export, 60-second cheat sheets, and Blitz Battle quizzes. It speaks 10 languages, including conversational Hinglish.

How we built it

FastAPI (Python) on Vercel serverless, Google Gemini (3.6-flash with a fallback chain so the tutor never goes down), Microsoft Edge Neural TTS + Web Speech for voice, Tesseract OCR for reading handwritten work, KaTeX for math rendering, and vanilla JS + Tailwind + Mermaid.js on the frontend. The Socratic behavior is pure prompt engineering — the hardest part was making the AI withhold answers and ask the right question instead.

Challenges we ran into

Three big ones. First, latency vs. interactivity: streaming voice that can be interrupted mid-sentence without glitching took real work on the Web Audio clocking. Second, math: getting LaTeX to render cleanly in chat and be spoken naturally ("x squared" not "x caret 2") needed a dedicated speech-cleaning layer. Third, zero-friction access — we learned that if a judge or student hits a login wall, you've already lost them, so guest mode had to work flawlessly.

Accomplishments that we're proud of

A live demo anyone can try in one click — no sign-up. Real two-way voice tutoring, an AI whiteboard that pinpoints the exact line where your logic broke, and 41 automated tests passing. It's fully open-source (MIT).

What we learned

The unglamorous lessons mattered most: graceful degradation (when the AI API fails, the app must degrade honestly, never hallucinate), language matching (an English question deserves an English answer), and that onboarding friction kills even the best product — every extra click before value costs you users.

What's next for ClearMind Pro

Teacher and parent dashboards, deeper syllabus integrations for competitive exams (JEE/NEET), offline PWA mode for low-connectivity students, and expanding beyond STEM into languages and humanities.

Problem Statement

Students re-read dense textbook paragraphs with near-zero comprehension, memorizing keywords instead of building mental models. When a student makes an error, standard chatbots dump a wall of text — never diagnosing where the misconception actually occurred. One-size-fits-all studying ignores how memory works: without spaced review and prerequisite mapping, knowledge decays within days.

Solution Overview

ClearMind Pro replaces passive reading with an active, empathetic learning loop. Its AI tutor Luna calibrates to each learner's academic tier (Class 6 → B.Tech, MBBS, self-directed), auto-generates their syllabus, and guides them through an active 6-stage mastery cycle in a dark-mode cockpit with voice synthesis, interactive chat, and live concept mapping.

Key Features

• Socratic Voice Coach — live two-way voice tutoring that asks guiding questions instead of spoon-feeding answers
• AI Smart Whiteboard — students draw or upload handwritten steps; AI marks with red/green pens to pinpoint the exact line where reasoning broke
• Generative HTML5 Labs — on-the-fly 60 FPS physics/chemistry sandboxes generated for any topic
• Feynman Reverse-Tutor Arena — students prove mastery by teaching Leo, a curious AI kid who asks trap questions and grades clarity
• 3D Flashcards + Anki Export — spaced-repetition decks (SM-2) with 3D flip physics and 1-click Anki export
• 60-second cheat sheets, prerequisite & mastery tree, Blitz Battle Arena quizzes
• Intuitive Deconstruction in 10 languages, including conversational Hinglish

Technologies Used

FastAPI (Python) on Vercel serverless, Google Gemini (gemini-3.6-flash with fallback chain), Microsoft Edge Neural TTS + Web Speech, Tesseract OCR, Vanilla JS + Tailwind CSS + Mermaid.js, hosted on Vercel.

Target Users

School students (Class 6+), college students, self-learners preparing for competitive exams — anyone who learns better by doing than by re-reading.

Built With

• edge-neural-tts
• fastapi
• google-gemini
• javascript
• mermaid.js
• python
• tailwind-css
• tesseract-ocr