LOGYRA (Think clearly. Learn deeply.)

LOGYRA is an AI-powered study workspace that helps students plan better, practice smarter, and learn from mistakes instead of answers.

Comment

🧠 LOGYRA: Learn Deeply, Think Clearly

Version AI-Powered License

Stop memorizing. Start reasoning. LOGYRA is an AI-powered study workspace designed to bridge the gap between knowing a formula and understanding the logic. Built for students who want to master their subjects, not just pass the test.

✨ Inspiration

Most AI tools today are "Answer Engines"—they give you the result, but they steal the learning process. As a student, I realized that getting fast answers leads to repeating the same mistakes during exams. LOGYRA was built to be a "Thinking Partner," focusing on how a student thinks rather than just what they answer.

🚀 Feature Overview

🎯 Core Product Features

  • 🔍 MistakeLens (Flagship): Upload a photo of your handwritten work. Instead of giving the answer, it acts like a human examiner, pointing out logical gaps and nudging you toward the right path.
  • 👨‍🏫 Tutor AI: Interactive, step-by-step explanations that adapt to your level of understanding.
  • 📊 Weakness Tracker: Analyzes your performance history to identify recurring conceptual gaps.
  • 📝 Revision AI: Automatically generates high-yield revision notes based on your most challenging topics.

🛠️ System Features

  • 📅 Timetable AI: Generates realistic, dynamic study plans based on your syllabus and available time.
  • ✍️ Exam AI: Create custom mock exams that mimic real test environments.
  • 🛡️ Study Coach: A motivational companion to help maintain consistency and focus.

🤝 Community & Social Features

  • 👥 Friend Connections: Connect with friends, send and accept friend requests, and build your personal study circle.
  • 💬 Community Discussions: Topic-based discussions with a structured message feed for doubts, strategies, and peer help.
  • 📨 Direct & Threaded Messages: Message-based interactions inside the community for focused conversations.
  • 🖼️ Profile Pictures & Identity: Custom profile images to personalize your presence and make interactions more human.
  • 🏷️ Public Profiles: View friends’ progress, activity, and shared learning milestones (privacy-controlled).

👤 User Management & Admin

  • 👤 Personal Profiles: Centralized dashboard for user details, progress, preferences, and social activity.
  • 🛠️ Admin Panel: Insights into user engagement, performance trends, and community health.
  • 📞 Contact & Support: A direct channel for users to send queries and for administrators to manage and respond efficiently.

💻 Tech Stack

Layer Technology
Frontend React TypeScript TailwindCSS Vite
Admin Panel React TypeScript Vite
Backend Node.js Express Pusher Nodemailer
Database MongoDB
AI Engine Gemini (Multimodal Reasoning & Tutoring)

🔧 Project Setup

1️⃣ Prerequisites

  • Node.js: v24.x or later.
  • Database: A MongoDB connection string (local or MongoDB Atlas).
  • Gemini API: A valid API Key from Google AI Studio.
  • Nodemailer: A Gmail account with an "App Password" enabled for sending emails.
  • Pusher: API credentials from a free Pusher account.

💡 Decisions & Assumptions

  • Prompt-Based AI Agents:
    Instead of hard-coding complex backend logic, I chose a prompt-engineering–driven architecture. Each major feature (e.g., MistakeLens, Tutor AI) behaves as a specialized AI agent with a defined persona and strict constraints — especially the rule to guide, not reveal answers.
  • Intentional Simplicity in Auth:
    Authentication and user profiles are intentionally lightweight in this phase to keep the focus on AI reasoning quality and user learning experience, rather than infrastructure complexity.
  • Multimodal-First Design:
    The system assumes students learn best through visuals. As a result, multimodal AI capabilities are core to MistakeLens, enabling meaningful analysis of handwritten work instead of plain text interpretation.
  • Data-Driven Feedback Only:
    All insights generated by the Weakness Tracker are derived strictly from actual user interactions and mistakes, ensuring feedback is grounded and trustworthy rather than synthetic or random.

🚧 Challenges & Learnings

  • The Hallucination Problem:
    Early iterations of MistakeLens occasionally identified mistakes that weren’t present. This was mitigated by enforcing structured reasoning and internal verification, ensuring the AI first solves the problem before evaluating the student’s approach.
  • Feature Scope Discipline:
    The original vision included a full-scale student social network. Through iteration, I learned that MistakeLens is the core differentiator, and community features should support learning rather than distract from it.
  • Designing AI as a Mentor:
    One key realization was that effective educational AI doesn’t aim to be the smartest — it aims to be the most pedagogically aware, asking the right questions at the right time to promote independent thinking.

🔮 What’s Next?

  • [ ] Basic Progress Tracking:
    Store solved questions, attempts, and timestamps to show simple improvement stats (weekly activity, accuracy trend).
  • [ ] File Upload Support (Images & PDFs):
    Allow users to upload notes or question PDFs and extract text for AI reference during explanations.
  • [ ] Saved AI Sessions:
    Let users revisit past MistakeLens or Tutor AI conversations for revision.
  • [ ] UI & UX Refinements:
    Better loading states, empty screens, and mobile responsiveness.

Build with ❤️ for the future of learning.

Built With

Share this project:

Updates