Edulens-ai

EduLens AI — Adaptive Learning Companion. Uses NLP to analyze student understanding from free-text responses, generates personalized explanations, and adapts learning paths in real time.

Comment

💡 Inspiration

The inspiration for EduLens AI came from a place of academic burnout. As a student, I’ve been there—staring at a screen, mindlessly clicking through multiple-choice questions just to finish an assignment.

I realized that this kind of "recognition-based" testing leads to shallow learning. You might recognize the right answer, but can you actually explain it to someone else?

I wanted to build something that moves beyond the "pick B for Mitochondria" approach and enforces a true "Explain-First" mindset.

🚀 What it does

EduLens AI is an adaptive learning platform that replaces traditional quizzes with a Cognitive Audit.

  • Explain-First Pedagogy
    Students explain concepts in their own words instead of selecting answers.

  • Bloom’s Taxonomy Classification
    The AI evaluates the depth of understanding—from basic remembering to advanced analysis.

  • Spaced Repetition
    I implemented the SM-2 algorithm to ensure long-term retention, not just short-term memorization.

  • Live Educator Dashboard
    Teachers get a real-time "God-view" of the classroom, including Bloom-level badges and misconception alerts as students type.

🛠️ How I built it

I designed the system with a high-performance, agentic architecture:

  • The Brain
    Powered by Llama 3.1-8B via Groq, enabling sub-2 second response latency for a conversational experience.

  • The Engine
    Built with React and TypeScript, featuring a glassmorphism UI and Framer Motion for smooth transitions.

  • The Logic
    Implemented the SuperMemo-2 (SM-2) algorithm for mastery tracking and adaptive learning.

  • Real-time Sync
    Used WebSockets (ws) to create a live bridge between students and educators with minimal latency.

🚧 Challenges I ran into

The biggest challenge was balancing hallucination vs. pedagogy.

Early versions of the AI were too lenient—it would assign high scores if students used correct keywords, even when their reasoning was incorrect (e.g., "Mitochondria creates oxygen").

To fix this, I built:

  • A Gatekeeper Agent to detect plagiarism and shallow responses
  • A Diagnostic Agent with strict weighted scoring

After multiple prompt iterations, I achieved 83.3% accuracy in misconception detection, which was a major breakthrough.

🏆 Accomplishments I'm proud of

  • Sub-2s Feedback Loop
    Achieved near-instant cognitive audits for a seamless user experience.

  • Interactive Knowledge Graph
    Built a 2D visualization that encourages students to explore dependencies and mastery paths.

  • Agentic Reliability
    My multi-agent system successfully filters out 56% of low-quality or plagiarized responses before grading.

🧠 What I learned

I learned that prompt engineering is essentially pedagogical engineering.

When building educational systems, you can’t just ask an LLM to "grade this"—you need to define the cognitive framework you're evaluating against.

I also realized that data validation is critical. Without a proper validation pipeline, it’s impossible to understand where the model fails until it impacts real users.

🚀 What's next for EduLens AI

  • Multi-modal Evidence Mode
    Allow students to explain concepts using diagrams or voice input.

  • Predictive Remediation
    Build systems that anticipate misconceptions based on a student’s learning trajectory before they occur.

Built With

Share this project:

Updates