AI Shadow Classroom : Silent background monitoring that detects student confusion and to personalize learning.

NeuralBits

Sankalp Dangi,Saniya Khatoon, Sakshi Budholiya(IPS academy)

Training / education / pedagogy

We are building an AI learning platform that works like a personal tutor inside a video player. It creates real-time notes and learns each student's study style. By tracking behavior like repeated rewinds, it detects when someone is stuck — pausing to explain concepts in a simpler way. It supports multiple regional Indian languages via Bhashini API. If a class is missed, it generates instant AI summaries to catch students up. Built for students who never had access to private tutoring.

Every classroom has students who are completely lost but will never raise their hand. Maybe the explanation was too fast, or maybe English isn't their first language. Whatever the reason, they fall behind — and nobody notices. This keeps happening until they give up. We've seen it. We've felt it. The tools that exist today are often made for students who already have advantages: good devices, strong English, and paid coaching. Our platform levels the playing field. It detects confusion before it becomes failure, explains things in the language the student is comfortable in, and makes sure a missed class never turns into a missed semester.

We are three CS students from Indore. We didn't come from big cities or fancy colleges. We've sat in classrooms where the lecture moved on while half the room was still stuck on the previous slide. Nobody said anything. That's the problem we want to fix. We joined this challenge because we want to build something that actually helps students like us — not just students who already have every advantage. We want to prove that a small team from a Tier-2 city can build something worth showing to the world.

We are building a web platform using React, FastAPI, SQLAlchemy and SQL. We have a custom HTML5 video player built in React that tracks every rewind, pause and skip a student makes during a lecture. This data gets sent to our FastAPI backend and stored in our SQL database. When a student rewinds the same clip three times, the system automatically pauses the video and generates a simpler explanation for that exact moment using an LLM. For notes, we run the lecture audio through Whisper to get a full transcript. That transcript goes into an LLM which structures it into clean readable notes. Every time a student edits their notes, we save those changes in the database against their profile. The system analyzes their editing pattern and the next time they open a lecture, their notes already come formatted in their preferred style. All notes, summaries and explanations pass through the Bhashini API which translates everything into the student's chosen regional language in real time. Students can switch languages anytime from their dashboard. After each lecture ends, our backend pulls that student's confusion data — which moments they rewound, which sections they spent most time on — and passes it to an LLM to generate a personalized quiz targeting exactly those weak points. If a student missed a class, Whisper transcribes the full lecture and the LLM generates a structured summary with key points and questions so they can catch up before the next session. All student data, notes, quiz results and watch history are stored and managed through SQLAlchemy.