661–670 of 1001 posts
Cooking up a NSFW filter for marginalia search. Pipeline so far has gone like this: * Use the search engine's API to query a bunch of depravity * Use qwen3.5 to label the search results and generate training data * Try to use fasttext to create a fast model * Get good results in theory but awful results in practice because it picks up weird features * Yolo implement a small neural net using hand selected input featur…
I have been using online courses and youtube forever and they all have converged on a similar format. Basically, adapt a textbook to slides and add voiceover. Sometimes they'll be animated. Not for everyone but I like it for passive learning.
Being a web developer I always thought video was a strange way to deliver this information - you can't even copy the text! Videos are also hard to make and heavy on bandwidth. So after iterating on different approaches to this over the last few years, I finally started on a new iteration called useful.
Currently on rev1, so early days: https://github.com/ohmstone/useful/tree/rev-1
There are a few more things I want to add to it but I want to get back to what I was doing (the bass guitar stuff). So I will make few of these website-as-video courses based on my projects to try and prove the concept.
Some of the nerdier features useful has:
- Uses state of the art CPU-based TTS with voice cloning, realistic enough to not be distracting
- Very simple markup language to create the visuals
- Extensible slide content with simple plugin system
- Full website export with complete SEO/social metadata
- Export is a PWA, so it caches nicely and can work offline
- Self-hostable
- Export is optimized for low bandwidth, so it loads way faster than a video and uses - Minimal dependencies
Beyond my own use-case I figured it might be useful for others creating courses. One stretch-goal would be for people to turn what they are learning via LLMs into low-bandwidth courses like this so we don't have people burning energy asking the same questions and watching the same 4K videos.
An all-in-one tool for structured data extraction with LLMs.
$ struktur extract —input doc.pdf —schema schema.json —model openai/gpt-5
- can prepare documents (PDF->text etc.) - run multiple different extraction strategies - runs a full agent loop for data extraction in-process using Pi agent and just-bash.dev. It can grep through large files for example.
I was sick of getting cross-eyed when looking at tables in raw markdown and was just running it locally. This weekend I realized it might be useful for others.
The goal was simple as possible UX. Open url, drag and drop or paste into wysiwyg -> very readable and editable markdown. No sign up, no tracking, no fuss.
Of note, if you copy from the richtext mode, it copies raw markdown. The inverse is done with paste.
Based on feedback, I am working on very optional cloud-sync for as cheap as I can make it.
- chef personalities generating interesting recipes every couple days
- the ability to save and edit these recipes to suit your needs/ingredients
- the ability to schedule weekly meal plan generations that take the inspiration content and give you a plan and shopping list for the week.
We had our first kid this year and I've been having more trouble getting things together for home cooked meals. This is my attempt to make it is frictionless as possible. I'm working on getting instacart API access so I can build out the cart for the meal plan automatically, at which point I'm hoping this is a one click confirmation a week to keep interesting food flowing. Works great for scheduling baby meals as well!
Im building https://trypixie.com to legally employ my 7 year old child, save on taxes and contribute to her Roth IRA. Im also building https://www.keepfiled.com , a microsaas to save emails (or email attachments) to google drive I almost forgot, I also built https://statphone.com - One emergency number that rings your whole family and breaks through DND. I love building. I built all these for myself. unfortunately I…