News for hackers. Stuff that matters.

mantegna

Re: on Saturday November 07, 2026 @01:11PM (#50406662)
Short backstory, Opus 5.5 appeared to be so great at 3D modeling that I had no choice but to make something fun with it

I was always interested in anatomy and wondered how things actually work, as biology lessons with 2D pictures are the least clear stuff out there, so I built Bodylang, an open-source app that lets you chat/model/track your own body throughout the day and through the vast variety of all the situations that could happen to you

It's a bit early and not all the systems are present, but I find it truly insightful and fun to play with. I actually used another app to make Bodylang a desktop image (interactive, called Splash) on my second monitor, so I'm constantly interacting with it in the background

The LLM underneath works via your existing Claude Code subscription and its generally very cheap, mostly on Haiku/Sonnet

Have fun!

Re:Tradeviction – Long or short startup conviction on Wednesday September 02, 2026 @12:01PM (#49538294)
Built for those who have ever wanted to short startups they dislike and long companies they believe are hidden gems. An artificial conviction market to express beliefs and see how they evolve over time, compete in groups, and throw a few hot takes. All the feedback and ideas are very much appreciated!
Re:Free and Simple Chess Analysis on Wednesday May 06, 2026 @04:02PM (#48041004)
experimentally built chess analysis app as a hobby project: g6chess.com

early alpha for now, with the main idea to keep it simple and easy to navigate for every player level (better use desktop for now)

to simplify the use, you can inject the "g6" before any chess.com or lichess.org game you just played and press enter. I own both g6chess.com and g6lichess.org domains so we'll get your redirect right to the analysis page built on top of my previous opensource involvements: ultrachess and ultrachess-React. explanation backend is closed-sourced for now, a bit too early for public, but planning to opensource it eventually too

main challenge is to make thoughtful formulas for move labeling, as it tends to overuse 'blunder' and 'best' when the settings are too rigid; and it's kinda tricky to train the LLM to explain the moves, even with comprehensive context. my personal impression is that taketaketake do it the best, but I spent just a few days for now, so we'll see where we end up

happy for any feedback and ideas:)

Re:Ask HN: What are you working on (August 2024)? on Sunday August 25, 2024 @05:48AM (#41345981)
Working on plump.ai – job-seeker's AI hiring agent. Think you have a personal career manager who searches for better job opportunities, applies you to them, speaks with recruiters on your behalf up until the interview scheduling, and does it 24/7 without involving you until real human action is required.