Media Matchmaker (Part II)

How did Media Matchmaker perform
LLMs
prompting
GPT Builder
ChatGPT
Author

Leesa

Published

March 23, 2024

image.png

Produced by ChatGPT

Media Matchmaker (Part II)

In the previous post, we explored how GPT Builder works and how to complete the initial set-up of a GPT. I created one called Media Matchmaker. It is designed to provide media (books, movies, tv, songs, podcasts, etc) based on users’ requests.

Let’s test how it works!

Testing out the system

The GPT Builder has a “Preview” panel, where you can test out the GPT before publishing it. In this tab, I prompted: “I just finished reading the sparrow. Can you recommend something that’s a similar vibe?”

It replied: image.png

These recommendations were pretty good! I am definitely interested in reading these books definitely, so that means that their recommednation system is working well.

Making some tweaks

After he initial configuration, it is still possible to tweak the recommendations based on how you feel about the preview. I noticed that in the response, the GPT only recommended books. I think it would be cool if the system either inquired about what type of media you are looking for, or offered different types. For example, if I just finished the Three Body Problem book, it should recommend that I now go watch the tv show on Netflix, as well as other similar science fiction books.

I asked the builder to make that change: “Can you update the GPT so that it asks what type of media the user is looking for? It recommended all books, but I may have been interested in TV or something else.”

It make the changes and the page looked like this: image.png

The GPT then provided a recommendations for books, tv shows, movies, podcasts, and songs. This worked very well.

Next I asked an even vaguer question to test the system: “I didn’t like the book Dare to know, but I like the idea. What would you recommend?”

It responded:

image.png

These were great recommendations!

Next I asked: “I have been really into renee rapp recently. what are some other artists / songs I could listen to like that?”

It responded:

image.png

I really like this recommendation system so far. It was super easy to build, it covers all of the functionality that I would need for now.