pdfGPT enables conversational interaction with PDF documents by leveraging GPT language models. It uses embeddings to represent PDF content and retrieves relevant passages to ground LLM responses. The project claims a simpler architecture than typical RAG pipelines by using embeddings without vector databases or explicit indexing. It supports multiple OpenAI models including GPT-3.5-turbo, text-davinci-003, and GPT-4, and includes features like chat history and pre-defined questions.