←back to thread

Ask HN: Most interesting tech you built for just yourself?

1901 points l2silver | 9 comments | 27 Apr 23 15:04 UTC | HN request time: 0.666s | source | bottom

Maybe you've created your own AR program for wearables that shows the definition of a word when you highlight it IRL, or you've built a personal calendar app for your family to display on a monitor in the kitchen. Whatever it is, I'd love to hear it.
1. hectormalot ◴[28 Apr 23 05:33 UTC] No.35737784[source]▶
A recipe manager for our family that strips all the SEO text out using the OpenAI API. I built this after someone in our family got diagnosed gluten intolerant and we had to make changes to our usual recipes.

Normal recipe sites tend to be full of irrelevant (SEO optimized) text, ads and tracking, and I wanted something to just get the recipe in a clean form.

It’s a basic web application (mostly in Go) to manage recipes. New recipes are imported from an URL, after which it extracts the plain text from the site and uses GPT to get a markdown formatted recipe and list of ingredients.

This would’ve been much harder pre-GPT, but now was trivial to implement.

replies(3): >>35737875 #>>35738004 #>>35743587 #
2. xnickb ◴[28 Apr 23 05:48 UTC] No.35737875[source]▶
That already exists as a browser extension afaik. Not sure about gluten part though
replies(1): >>35737961 #
3. hectormalot ◴[28 Apr 23 06:00 UTC] No.35737961[source]▶
Oh for sure. I think Paprika (?) does a decent job. To be fair, it started as a learning project for myself.

As a minor detail I also translate everything to the same language as part of the transformation. Just a bit of prompt experimentation.

4. iaaan ◴[28 Apr 23 06:06 UTC] No.35738004[source]▶
The mobile app Paprika does this and works great, in case anyone wants to use something like this.
replies(2): >>35739717 #>>35745547 #
5. IanCal ◴[28 Apr 23 10:57 UTC] No.35739717[source]▶
Paprika is excellent, it's also something you pay for and just get the app - no subscription.
replies(1): >>35747859 #
6. Paul-Craft ◴[28 Apr 23 16:29 UTC] No.35743587[source]▶
> This would’ve been much harder pre-GPT, but now was trivial to implement.

I wonder how close looking for a group of lines that start with a number, then taking all the text following that group of lines would get you. I bet that would get you pretty close to the desired effect most of the time.

I also wonder if more standard NLP methods might work here, rather than using the full power of an LLM. Instructions are grammatically constructed as commands, so if you start with doing what I mentioned in the previous paragraph, then parse each sentence following the group of lines starting with numbers, you should be able to determine which ones are commands. If a paragraph contains no commands, it's obviously not directions.

Just some random thoughts. I do see what you mean, though: definitely not trivial sans assistance from an LLM.

replies(1): >>35750231 #
7. macrael ◴[28 Apr 23 18:55 UTC] No.35745547[source]▶
I love Paprika, one of my favorite apps I use. It even syncs a grocery list between my phone and my Mac.
8. suddenclarity ◴[28 Apr 23 22:22 UTC] No.35747859{3}[source]▶
On the downside, you'll need to purchase each version separately. I bought it on Apple and then moved to Windows. I'm considering buying it again but I'm curious how well Obsidian would work for it considering it's free and has the rest of my life.
9. hectormalot ◴[29 Apr 23 05:12 UTC] No.35750231[source]▶
Was thinking the same thing initially, but it would have been quite a time investment. In this case I just make an API call and it’s done.

I prompt GPT to return what I want in JSON format and that works surprisingly well, so it turned out easy to parse too.