Vidya — AI Knowledge Companion

An AI-powered learning operating system that transforms any textbook into 7 interactive learning modes — powered by Amazon Nova Lite, Embed, and Vision.

Comment

Inspiration

Students in 2026 don't fail because information is missing — they fail because there's too much of it. Google, YouTube, PDFs, lecture slides — all explaining the same concept differently. The result? Cognitive overload, procrastination, and zero deep learning.

Coaching centers charge thousands and rush through syllabi. Slow learners get left behind. Professors lose students after 8 minutes. And when students try to self-study? They open YouTube "to learn" and end up watching shorts for 3 hours.

We asked: What if AI didn't just answer questions — but structured the entire learning journey around how students actually behave?

That's Vidya (विद्या — Sanskrit for "knowledge").

What it does

Vidya transforms any textbook into a complete, interactive learning experience with 7 powerful modes — all powered by Amazon Nova:

  1. Subject Map — Enter any book. Nova Lite extracts every chapter, every topic, tagged by difficulty and read time. A complete roadmap — no guessing.
  2. Deep Concept — First-principles explanations with live Concept Networks. Not Wikipedia summaries — content designed to hold attention.
  3. Quick Revision — Sharp, exam-ready notes generated in seconds.
  4. Exam Mode — Focused content on exactly what examiners test.
  5. Knowledge Graph — A D3.js force-directed graph showing how concepts interconnect. See the big picture the way experts think.
  6. Doubt Engine — Ask anything, get instant Nova-powered answers. No waiting for office hours.
  7. Reels — 60-second TikTok-style micro-learning cards. Bite-sized retention.
  8. Battle Mode — Compete against AI on any topic. Gamified, competitive, addictive.
  9. Scan Page (Nova Vision) — Photograph any textbook page, handwritten notes, or problem set. Nova Vision reads it, identifies the topic, and gives instant analysis. Physical paper becomes interactive learning.
  10. Focus Mode — Gamified Pomodoro timer with anti-sleep detection. No more "I'll do it tomorrow."

One subject. Seven+ ways to master it. Zero switching between apps.

How we built it

Backend: Python FastAPI with a modular agent architecture — each AI capability (content generation, doubt resolution, knowledge graphs, reels, battle questions, vision analysis) is a dedicated agent with its own prompt engineering.

Frontend: Vanilla JavaScript SPA with modular CSS architecture — 11 JS modules and 16 CSS files organized by feature. D3.js for knowledge graph visualization.

Amazon Nova Integration:

  • Nova Lite (nova-2-lite-v1) — Powers content generation, doubt resolution, knowledge graphs, reels, battle questions, and exam preparation across all learning modes
  • Nova Embed (nova-embed-v1) — Enables semantic search and cosine-similarity-based topic connections in the knowledge graph
  • Nova Vision — Reads and analyzes physical documents — textbook pages, handwritten notes, printed problem sets — turning paper into interactive learning

All three models work together through an orchestrator pattern, with streaming support for real-time content delivery.

Deployment: Render.com with the backend serving both API and frontend via FastAPI StaticFiles middleware.

Challenges we faced

  • Prompt engineering at scale — Each of the 7 learning modes required distinct prompt strategies. Exam Mode needs precision; Reels need brevity and hooks; Deep Concept needs depth. Balancing all within Nova Lite's capabilities took extensive iteration.
  • Streaming architecture — Real-time content streaming via SSE required careful handling of JSON extraction from partial responses.
  • Knowledge Graph coherence — Making the D3.js force graph produce meaningful, non-overlapping concept maps from Nova's output required custom cosine similarity thresholds using Nova Embed.
  • Vision pipeline — Handling diverse input formats (photos, PDFs, handwritten notes) through Nova Vision while maintaining consistent analysis quality.

What we learned

Amazon Nova's model family is remarkably versatile. Using Lite, Embed, and Vision together unlocked capabilities we didn't initially plan — like semantic search powering the knowledge graph, or Vision turning any physical page into a learning moment. The models complement each other in ways that made the whole system greater than the sum of its parts.

What's next

  • Multi-language support leveraging Nova's multilingual capabilities
  • Collaborative study rooms with shared knowledge graphs
  • Spaced repetition powered by Nova Embed similarity scoring
  • Mobile-native app with offline scan capabilities

Built With

Share this project:

Updates

Submission history