Inspiration

Students struggle with personalized, offline study help. Generic AI chatbots lack context about their courses, don't adapt to their learning style, and require internet. We built LynxMind – a complete offline AI study coach that works anywhere, even without Wi-Fi.

What it does

LynxMind uses three specialized AI agents:

  • Tutor – answers questions using RAG from uploaded PDFs
  • Quizzer – generates adaptive quizzes (easy/medium/hard) and tracks weak topics
  • Motivator – gives personalized encouragement based on your performance

Plus 17 more features: gamification (XP/levels), code runner, voice input/output, dark mode, Pomodoro timer, PDF study guide export, collaboration, analytics dashboard, spaced repetition, LaTeX math rendering, personality slider, and more.

How we built it

  • Local LLM: Ollama + llama3.2:3b (runs entirely offline)
  • RAG: FAISS vector database + sentence-transformers embeddings
  • Frontend: Streamlit with 8 custom tabs
  • Persistence: JSON files for user progress, XP, and review schedules
  • Libraries: PyPDF2, reportlab, matplotlib, plotly, pyttsx3, langchain

Challenges we faced

  • Making all 20 features work together without breaking the local LLM context
  • Ensuring the Pomodoro timer didn't block Streamlit's event loop
  • Generating diagrams from natural language descriptions
  • Implementing voice input within Streamlit's component architecture

What we learned

  • Offline AI is feasible and powerful for education
  • Multi-agent systems can create richer user experiences than single chatbots
  • Gamification dramatically increases engagement

Impact

LynxMind democratizes AI tutoring – no internet, no API costs, no data privacy concerns. Perfect for rural schools, low-bandwidth areas, and students who prefer local tools.

Built With

  • langchain-text-splitters
  • matplotlib
  • pandas
  • plotly
  • pypdf2
  • python
  • pyttsx3
  • reportlab
  • sentence-transformers
Share this project:

Updates