Classroom AI

this page is for getting summaries of the textbook and ask doubts to the chat bot. Also we can integrate open ai for online using API.
in this page we can provide audios and convert them into a notes
this page is for convert the lecture notes into Quiz for IQ tests

Inspiration

In many regional and state-board educational ecosystems, students face a significant digital divide. While generic AI tools excel at processing standard English materials, they often struggle with regional syllabus structures, mixed-language nuances (such as Tamil and English code-mixing), and complex textbook layouts. Furthermore, high-bandwidth internet isn't always reliable in every rural or semi-urban classroom.

We were inspired to build Classroom AI—a lightweight, localized, and offline-first study companion designed to equalize access to personalized learning. Our goal was to create a tool that turns any textbook PDF or recorded classroom lecture into bite-sized summaries, structured notes, and self-assessment quizzes without complex setup.

What it does

Classroom AI acts as a localized multimodal study assistant for students and educators:

Syllabus Processing & Summarization: Converts dense textbook chapters into clear, accessible study guides tailored for regional language contexts.

Automated Assessment Engine: Generates interactive practice quizzes, multiple-choice questions (MCQs), flashcards, and instant step-by-step explanations directly from study materials.

Lecture-to-Notes Transcriber: Processes recorded audio from classroom lectures into structured, searchable bulleted study notes.

Offline-First Retrieval: Uses lightweight vector indexing to allow continuous study access without relying on high-speed cloud connections.

How we built it

We engineered Classroom AI using a modular, low-latency processing architecture:

Document Engine & Retrieval: Uploaded state-board PDFs are parsed and split into semantic chunks. We built a local Retrieval-Augmented Generation (RAG) pipeline using LangChain and ChromaDB / FAISS vector storage to index concepts locally.

Localization & Summarization Layer: We integrated fine-tuned multilingual NLP models (IndicBERT / multilingual LLMs) to extract core ideas and output contextual explanations in clear regional language syntax.

Audio Processing Pipeline: Recorded classroom lectures are processed through OpenAI Whisper for high-accuracy speech-to-text conversion, followed by automated NLP extraction to yield bulleted takeaways.

User Interface: Built a clean, intuitive dashboard using Streamlit / React allowing students and teachers to drag-and-drop materials and interact with generated study modules seamlessly.

Challenges we ran into

Handling Complex PDF Layouts: State-board textbooks often contain multi-column text, embedded diagrams, and non-standard Tamil/English font encodings. Ensuring clean text extraction required fine-tuning our PDF parsing and chunking rules.

Context Retention in Regional Speech: Local classroom audio recordings frequently feature regional accent variations and ambient background noise. Optimizing Whisper parameters and post-processing text tokens was essential to maintain high transcription accuracy.

Balancing Latency & Model Performance: Running local vector search and LLM inference quickly on standard hardware required optimizing our chunk sizes and embedding dimensions to maintain real-time responsiveness.

Accomplishments that we're proud of

End-to-End Pipeline Integration: Successfully connected document parsing, speech transcription, vector indexing, and quiz generation into a single continuous workflow.

High Regional Accuracy: Achieved clear, context-aware summarization that respects local state-board syllabus structures and language nuances.

Offline Capability: Delivered a functional prototype that operates without continuous high-bandwidth internet dependency.

What we learned

Effective RAG Design: Designing precise semantic chunking strategies dramatically improves the accuracy and relevance of generated study materials.

Accessibility-First UX: Keeping the user interface minimal and scannable makes complex NLP tools immediately intuitive for both students and educators.

Multimodal Orchestration: Chaining audio transcription, semantic vector search, and automated text generation provided invaluable hands-on experience in modern AI workflow design.

What's next for Classroom AI

WhatsApp & Low-Bandwidth Bot Integration: Deploying a WhatsApp interface so rural students can send voice notes or text prompts to receive study summaries over simple mobile data.

Multi-Language Expansion: Expanding native fine-tuning support to additional regional Indian languages beyond Tamil.

Interactive Analytics Dashboard: Adding progress-tracking dashboards for teachers to view class performance based on quiz results generated by the system.