ShikshaAI

“Where every student gets a personal tutor — even when teachers aren’t around.”

Comment

Inspiration

We noticed a recurring problem in schools — especially government and rural ones: students struggle with doubts after class hours, teachers are overloaded, and existing ed-tech apps are expensive, generic, or not aligned with CBSE/State Board syllabi. During conversations with juniors preparing for boards, we saw them turning to random YouTube videos, unreliable AI responses, or guessing through problems. Teachers also shared how they wished they had tools to track student understanding and generate practice material automatically. That sparked our idea — an AI tutor that works like a real mentor, aligned with our textbooks, supports teachers, and works even with poor internet.

What it does

Our system is an AI-powered learning assistant for class 9–12 students. It allows students to:

  1. Ask conceptual questions (e.g., “Explain Ohm’s Law”)
  2. Get syllabus-aligned explanations referenced to NCERT chapters
  3. Solve numericals step-by-step, revealing hints gradually
  4. Practice MCQs based on difficulty level
  5. Track weak topics and get recommendations
  6. Generate 7-day revision plans automatically

Teachers get:

  1. Class-wise performance analytics
  2. Topic heatmaps
  3. Auto-generated worksheets, PYQs, and model answers
  4. Ability to edit or upload questions It will be built as a web-based PWA, supports English + Hindi, and works offline for schools with poor connectivity.

How we will build it

We will develop the system using:

  1. Frontend -> React PWA to allow app-like experience with offline caching
  2. Backend -> Python FastAPI / Node.js to handle logic, routing, and student interactions
  3. Database -> PostgreSQL to store syllabus, PYQs, questions, and student profiles -> Local SQLite caching for offline school deployments
  4. AI Engine -> A Retrieval-Augmented Generation (RAG) system -> Detect chapter/topic -> fetch syllabus content -> Rewrite student-friendly explanations with an LLM
  5. Math Engine -> Symbolic solving templates to generate stepwise hints
  6. Recommendation Engine -> Track correctness per topic -> Identify weak areas -> Adjust practice difficulty
  7. Teacher Dashboard -> Analytics charts -> Heatmaps for weaknesses -> Content review and upload tools
  8. Offline Functionality -> PWA service worker caching -> Optional mini edge server (Raspberry Pi) for local AI + syncing

Challenges we expect to face

  1. Ensuring factual accuracy -> LLMs hallucinate, so we will anchor answers to syllabus notes.
  2. Designing step-by-step hint structure -> We expect iterations to make it intuitive for students.
  3. Implementing offline-first functionality -> Caching, conflict management, and syncing will require careful engineering.
  4. Teacher acceptance -> We will build feedback loops — allowing teachers to review, edit, and upload content.

Accomplishments that we're proud of (so far)

  1. A well-researched system design
  2. Syllabus-mapped content framework
  3. Strong pedagogical principles (mastery learning + scaffolding)
  4. A teacher-support mindset rather than AI replacement
  5. Architecture planning

What we learned

  1. AI in education must be curriculum-aware, not generic
  2. Students value stepwise guidance more than ready-made answers
  3. Teachers want AI that assists, not threatens them
  4. Rural deployments need offline-first design
  5. Personalization requires data + learning science, not just models

What's next for "ShikshaAI"

  1. Build student and teacher MVP interfaces
  2. Implement retrieval + syllabus mapping
  3. Deploy stepwise solver for math
  4. Add learning analytics and adaptive practice
  5. Integrate Hindi/English explanations
  6. Run pilot tests with sample chapters
  7. Voice-based interaction + misconception detection
  8. Full test-paper generator
  9. Scale to more subjects, more languages, and Indian boards

Long term vision

Make every student feel guided, every teacher feel supported, and every school become digitally empowered — no matter where it is.

Built With

Share this project:

Updates