Setting up a private local LLM for document summarization without cloud dependency

I have a collection of sensitive PDF documents I need to process regularly. I want to use a Large Language Model for summarization and Q&A, but I cannot upload this data to cloud-based APIs due to privacy policies. I have a PC with a mid-range GPU (RTX 3060 12GB). What is the most efficient setup to run an open-source model like Llama 3 locally? I need a recommendation for the specific model size that fits within the memory constraints while still understanding complex documents. Are there lightweight GUI interfaces available that integrate with local PDFs without requiring command-line coding? I prefer solutions that ensure the processing happens entirely offline. Please focus on stability and user experience for a non-programmer.