SnapClass

SnapClass is an advanced on-device edge AI solution designed for low-connectivity, high-density classroom environments. Powered by Snapdragon’s Hexagon NPU, it runs open-source large language models (LLMs), image captioning, and audio transcription entirely offline. SnapClass automates personalized learning by transcribing lectures, analyzing textbook content, and generating adaptive quizzes—bridging educational gaps in underserved regions where internet access and qualified educators are limited.

App architecture

Models

• Whisper-small (242M params) via "openai/whisper-small"
• nougat-small (247M params) via "facebook/nougat-small"
• blip-image-captioning-base via "Salesforce/blip-image-captioning-base" running parallelly with whisper-small
• Phi-3.5-mini-instruct (3.82B params) via "AnythingLLM" running locally via ONNX accelerated by Snapdragon's X Elite's NPU

Features

• 🖼Teacher dashboard with file upload and analytics view
• Lecture and textbook PDF/audio upload
• AI-based question and answer evaluation
• Uses Both CPU and NPU for faster on-device processing
• Identifies weak syllabus topics per student or group
• Fully functional offline – no internet needed
• Lightweight and fast inference using sentence embeddings

Setup & Usage

Step 1: Setup Local Hotspot (No Internet)

We use MyPublicWifi to create a local area dead network

1. Download and Install MyPublicWifi.
2. Open the app, set:
   • Network Access = No Internet Sharing
   • Turn on hotspot
3. Note the IP address shown in the app. This IP will be used to access the server from other devices.

Step 2:Install Python Dependencies

1. Clone this repository

git clone https://github.com/Abbilaash/SnapClass.git
cd SnapClass

2. Install requirements

pip install -r requirements.txt

Install AnythingLLM and activate AnythingLLM NPU to process LLM models in Qualcomm Hexagon NPU. Download Phi 3.5 Mini Instruct 4K 2.00GB model. Get you AnythingLLM Developer API from Settings>Tools>Developer API>Generate New API Key and get you workspace slug. Replace you API key in server/config.yaml as api_key: <API> and workspace_slug: <my-workspace>

Step 3: Start the server

cd server
python app.py

Once started, the server runs locally on port 5000

Access the Web Interface

• Teacher Dashboard (same server device) http://127.0.0.1:5000/admin Use the IP address shown in MyPublicWifi ad access the students portal
• Student Test Page http://<IP>:5000/test Replace with the one shown in the MyPublicWifi app after enabling the hotspot

Screenshots

Authors

A T Abbilaash - [email protected]
Nivashini N - [email protected]