This is a submission for the DEV Weekend Challenge: Community
The Community
I grew up watching teachers struggle in classrooms where the internet is either absent or unreliable. In rural Area, millions of school teachers face the same problem every single day: they are expected to deliver quality, engaging lessons — but the digital tools that could help them are locked behind a stable internet connection they simply don't have.
Free AI tools like ChatGPT and Gemini exist — but they're completely useless when there's no internet. Offline AI tools like Ollama and LM Studio exist — but they require terminal commands, admin rights, and software installation that a government school teacher cannot do.
What I Built
PathShala AI is a browser-native, zero-install AI teaching assistant that runs 100% offline after a one-time setup — no app, no admin rights, no internet, no API key. Ever.
A teacher downloads one HTML file. Opens it in Chrome. Done. From that point, the full power of a local LLM is available — in a classroom with no WiFi, in a village with no mobile data, even on a flight.
Why a Single HTML File?
Rural school computers run older Windows with no package manager, no Node.js, and no IT support. A single HTML file that a teacher can open in Chrome is the only deployment format that works in this reality.
Installation is: download file → double-click. That's it.
Demo
Code
How I Built It
Tech Stack:
WebLLM (Apache TVM) – In-browser inference
Llama 3.2 1B (q4f32_1-MLC) – Quantized model
WebGPU – Hardware acceleration
Cache API + IndexedDB – Full offline storage (~900MB)
Web Speech API – Offline TTS
Vanilla HTML/CSS/JS – Zero dependencies
Architecture
Query → Local Model (Cache) → WebGPU Inference → Text/Voice Output
No backend. No API key. No data leaves the device.
Team Submissions: Build By Malathi Venkatesan
username: malathilatha199615
Top comments (0)