What is this?
The Office Oracle AI chatbot is a complete AI chatbot built on top of langchain, Pinecone.io and OpenAI's GPT-3 model. It demonstrates how you can build a fully-featured chat-GPT-like experience for yourself, to produce an AI chatbot with any identity, who can answer factually for any arbitrary corpus of knowledge.
For the purposes of demonstration, I used the popular Office television series, but this same stack and approach will work for AI chatbots who can answer for a company's documentation, or specific processes, products, policies and more.
Be sure to check out my three-part video series on YouTube, where I break down the entire app end to end, and discuss the Jupyter notebooks and data science elements, in addition to the Vercel ai-chatbot template I used and modified for this project:
Intro video and demo
Jupyter notebooks deep-dive
Next.js vercel template ai-chatbot deep-dive
Open source code
I open sourced the Jupyter notebooks that I used to prepare, sanitize and spot-check my data here:
The data workbench notebook handles fetching, parsing and writing the data to local files, as well as converting the text to embeddings and upsertings the vectors into the Pinecone.io vector database.
The test workbench notebook demonstrates how to create a streamlined test harness that allows you to spot check and tweak your data model without requiring significant development changes to your application layer.
I also open sourced the next.js application itself