
Pathshala AI is an AI-powered bilingual tutoring platform designed to improve primary education accessibility in rural Nepal. In many rural communities, students study using Nepali-medium textbooks but often face limited access to experienced teachers, personalized tutoring, and high-quality educational resources. This creates learning gaps, especially in subjects such as science and mathematics. Pathshala AI was designed to help bridge this challenge through AI-assisted educational support. The system combines Retrieval-Augmented Generation (RAG), multilingual workflows, and AMD-powered inference infrastructure to create textbook-grounded educational explanations. Instead of relying only on local textbook content, the platform leverages multiple English educational resources internally to strengthen conceptual understanding and then adapts explanations into simpler, understandable Nepali for students. Students can interact naturally using: • English • Nepali • Romanized Nepali (e.g., “friction vaneko ke ho?”) The platform retrieves relevant educational context, generates simplified explanations, creates adaptive quizzes for reinforcement, and provides parent/teacher learning summaries to support progress tracking. Technically, the system was built using: • AMD MI300X infrastructure • vLLM for scalable inference • Qdrant vector database for retrieval • FastAPI backend services • Multilingual RAG workflows Pathshala AI is not designed to replace teachers, but to provide scalable educational support where tutoring and learning resources are limited. Future plans include voice tutoring, WhatsApp integration, multilingual subject agents, and wider deployment for low-resource educational environments. The project was built independently during the AMD Developer Hackathon while proudly representing Nepal among teams from around the world.
10 May 2026