Jetson Orin Nano RAG

Description:

Jetson-Orin-Nano-RAG is a fully offline retrieval-augmented generation system built for the NVIDIA Jetson Orin Nano that lets users upload PDFs, create a local searchable knowledge base, and ask natural-language questions with cited answers. The project combines PDF text extraction with OCR fallback for scanned documents, generates embeddings for document chunks, stores them in a FAISS vector index, and uses a local Qwen2.5 GGUF model through llama.cpp for on-device response generation. It also includes practical features like incremental indexing with file hashing, metadata tracking, MMR-based retrieval for better context selection, and a Streamlit interface for document upload, querying, and source inspection, making it a strong edge-AI project that demonstrates local LLM deployment, information retrieval, and GPU-accelerated inference without cloud dependencies.