Tutorial: How to Build a Customer Support Discord Bot with Completions and Embeddings APIs

[glad shoal]

Hey all, a couple of weeks ago we launched https://discord-gpt.com/ -- a template that makes it really easy to get your own GPT-powered Discord bot up and running. We've had a ton of requests from people looking for customer support bots that can connect to their own knowledge base, so we thought we'd write a tutorial.

https://autocode.com/guides/how-to-build-a-gpt-support-discord-bot/

• This guide will show you how to generate great completions
• ... and teach you a bit about embeddings
• ... and show you how to connect OpenAI to Discord
• ... and show you how you can power a knowledge base via Google Sheets

It's meant to be approachable to a non-developer or first-time developer audience as well, so I hope y'all don't mind if it holds your hand a little! Let me know if there are any questions.

• Keith

[sudden ice]

@glad shoal I read your complete blog and was really very helpful. Thanks!

[glad shoal]

No problem! Glad I could help. 🙂

[sudden ice]

@glad shoal One question:

How can I search through a knowledge base if we let's say just have answers or simply want to search through a blog post and get answers based on user query?

In this case we won't be able to classify user query as there is no label (mapping of question and answers) and if we just simply compare user query to the blog post then I believe there will be multiple lines inside of blog post that'll match the user query

@glad shoal I'm thinking of multiple approaches

1. We can create labels first and then perform sentiment analysis, just like in the blog post
2. Divide the blog post into sentences and then compare embeddings of user query and each sentence query. After that I can rank them and give it to completions API

Can you tell me which approach will be beneficial or is there any better approach?