Papers GPT

Papers GPT is a personal demo project by software engineer and entrepreneur Jesse Zhang, built to show how large language models can be given deep, specialized knowledge by feeding them external documents. Rather than being a polished product, it is a proof-of-concept for connecting arbitrary data sources to GPT so it can answer detailed questions about specific material. The demo works by downloading source documents, extracting their text, and indexing them with embeddings before feeding relevant chunks into GPT to generate answers. It is built on the GPT Index library and runs on a small Flask server. Supported data sources go beyond scientific papers to include Google Docs, GitHub repositories, AI and quantum computing reference material, DeepMind reinforcement-learning papers, and author-specific searches such as Andrej Karpathy's work. The interface is a terminal-styled Q&A window where users type natural-language questions about the selected data source. To use the tool, visitors must supply their own OpenAI API key with a set_key command; the creator recommends generating a temporary key and deleting it afterward for safety. The tool is not optimized for mobile screens, and because it is a side project rather than a maintained product, it has at times been disabled due to high traffic volume.