James Gurney
JG with PencilTagline

This site aims to give people an easy way to learn from James Gurney by asking questions.

It was made by Dan Gurney

It is built around the ChatGPT Retrieval Plugin, an open source project from OpenAI which is a Python server that connects to a Pinecone vector database. You can feed it written material, and it will categorize the text using vector embeddings, which are essentially a series of numbers that assign unique meaning to the text.

Here is a podcast episode with a great explanation of vectors if you're curious.

The cool part is that once all your material is categorized with vector embeddings, you can take any question and find the material that is most similar. It is much more powerful and precise than a simple string search, since it will search on the underlying meaning of the query.

Given this backend server, I then built a simple chat UI with NextJS. It takes the chats in your current session, finds the most relevant material in the vector DB, and then sends it all as a prompt to the ChatGPT API. The prompt is basically: "Imagine you are James Gurney. Given these chats and this background material, answer the user's question as best you can. You can infer from the material but don't make anything up."

So far the database has been populated with GurneyJourney blog posts, transcripts of instructional videos like the "In The Wild" series, and magazine interviews.

Over time we'll add more material to the database. We may also explore making this available as a ChatGPT plugin.

The code is open sourced here.