PathSala Offline AI Learning for Students in Low-Connectivity Regions
PathSala
PathSala is an offline-first Android AI tutor built for Nepali students. It runs a fine-tuned model.litertlm model directly on-device with Google AI Edge LiteRT-LM, so students can get educational support without depending on a continuous internet connection.
Key Features
Offline AI tutoring: Runs the local model.litertlm engine on-device with no cloud API dependency.
English and Nepali support: Lets students learn in the language they are most comfortable with.
Class-aware explanations: Adjusts tone and depth based on the selected class level.
Subject-focused learning: Supports Science, Math, English, Nepali, Social Studies, and Computer.
Student learning tools: Includes daily streaks, study planning, progress tracking, and generated quizzes.
File-based help: Lets students attach text files and ask questions about the document content.
Follow-up responses: Pathsala remembers past messages to keep the conversation continuous and context-aware.
Project Links
Demo Video: YouTube Demo
APK + Model Files: Google Drive Folder
Public Repository: GitHub Repository
Model Fine-Tuning Notebook: Kaggle Notebook
Installation & Setup
PathSala needs two files:
app-release.apk
model.litertlm
Both files are available in the Google Drive Folder.
Install the app
Download and install app-release.apk on the Android device.
Add the model
Download model.litertlm.
Keep the file name exactly as model.litertlm.
Create a new folder on the phone.
Move model.litertlm into that new folder.
Open PathSala and tap Choose model folder.
Select the folder that contains model.litertlm.
After the model is selected, PathSala starts the offline AI tutor from that local model file.
Architecture
React Native frontend: Handles UI, chat flow, subject selection, local state, and student-facing interactions.
Native Android bridge: Uses Kotlin to bind React Native with the local LiteRT-LM runtime.
On-device inference: Loads model.litertlm locally and streams responses from the Android native layer.
Local persistence: Stores profile, progress, selected model folder, and chat state on the device.
User Experience
With PathSala, a student can:
Ask questions in Nepali or English.
Select a subject and get focused support.
Solve academic doubts step by step.
Generate quizzes for practice.
Track streaks and study consistency.
Follow a revision or study plan.
Learn in low-connectivity environments.
Current Capabilities
Subject-focused tutoring
Bilingual question answering
Quiz generation
Study planning and revision support
Streak and progress features
File-based question answering
Fully offline local inference with LiteRT-LM
Follow-up responses