AI Analyzed My Bookshelf to Give Me Personalized Recommendations

AI Analyzed My Bookshelf to Give Me Personalized Recommendations

I took a video of my bookshelf and got AI to create a list of custom book recommendations for me.

This was a multi-stage project, but I’m really happy with the results. I’ll cover how I got there in detail, but here’s the cliff’s notes:

  1. Have Google Gemini identify the books on the shelf using the video
  2. Have Gemini clean the data that it just created
  3. Ask Gemini to give me recommendations based off the cleaned data

1. Identifying books

I passed Google Gemini the video of my bookshelf and tasked it to identify the books and output it in a machine-friendly format(JSON)

I didn’t move anything around on the shelf because I wanted as realistic results as possible.

It did way better than I expected.

It still hallucinated some (I don’t own a copy of Skyward, for example), but it got most of the titles.

2. Cleaning the data

I then fed it back into Gemini to clean the data a bit.

Mostly, it removed duplicates and titles that wouldn’t make much sense. In this specific run, this step was a bit superfluous, but it depends a lot on how bad the identification went.

3. Getting the results

Lastly, I fed the cleaned data back into Gemini and had it give me good recommendations based on what it saw.

It did a good job! A lot of these I have already read and just don’t have on my shelf, including more obscure ones like The Goblin Emperor, so it did a good job of figuring out my taste.

I want to hook this up to some sort of semantic index or something so that it can get descriptions for the books themselves, but that will have to come in a future version. Follow me on Twitter and LinkedIn for more demos like this!

Recent Posts