@CJHandmer this is the best part of having ai. it makes it so easy to chase little side quests that would've been left unexplored otherwise
English
Rittik Basu
258 posts
About 7 years ago I vacationed in Kauai and being a giant nerd, wondered how many waterfalls there are. I went home and wrote some neat Mathematica code and counted them all. Today, I wondered about that project, but not having access to the data, I simply prompted Claude Code to do it all from scratch. It took me about 2 minutes to kick this off, and it did much more comprehensive data analysis and graphing that I ever had time to bother with the first time.
Casey Handmer@CJHandmer
This has at least 1000xed my productivity on weird side projects. I have a very very long backlog that I'm clearing with alarming speed. Unbelievable that we used to fill buildings with smart people and have them type algorithms into computers letter by letter.
English
also don't worry this isn't vibecoded slop and won't be a cpu hog. in my testing it stayed under 1% cpu while active.
github.com/rittikbasu/luc…
English
loved this idea from @KunalKSavita but i wanted it to work everywhere and not just the browser.
so i built lucent.
it's a tiny macos menu bar app that syncs your wiz lights to any screen or app window on your mac.
Goonal@KunalKSavita
i built something :) today, i’m announcing aura. sync your smart bulb’s color to the vibe of your music. and all open source.
English
@aravindballa @OpenAI basically instead of asking me for every command codex would send it to a guardian model first and if it looked safe it got auto approved. so i didn’t have to click approve 500 times.
English
@OpenAI where did my guardian approval go 😭
i got this feature a week ago and it was perfect. guardian would approve safe commands so i didn’t have to give codex full access for long tasks. updated codex and it was gone.
please bring it back before full access nukes my mac
English
agent browser is so good!
a few months ago i was scraping a site and it took me a few days to get the data reliably. this time i just gave codex the prompt, it used agent browser, found there's an internal api we can use instead of scraping the dom and got all the data.
English
i left codex running before i went to sleep and this is what i woke up to🤯
guess the inference cost for this run.
English
English
English
built this 2 years ago, opened it last week and it was broken. the api it used died. so i spent the last few days fixing it. new api, more features & some easter eggs.
this is probably the fastest f1 client on the internet rn but don't quote me on that.
formulator.rittik.io
Rittik Basu@_rittik
Built a formula 1 client inspired by my newfound love for F1 racing. Stay updated with all the stats, scores, and standings in a beautifully designed UI that's as fast as your favourite team on race day! Check it out → formu1ator.vercel.app #buildinpublic #f1
English
just opened the repo randomly today and saw it has 23 stars?? thanks y’all!
Rittik Basu@_rittik
built a tiny (~1mb) cli tool to schedule prompts for claude code. so now when you hit the rate limit you can just schedule a prompt to auto resume your work. because the work is mysterious & important and must go on. try it: brew install --cask rittikbasu/wakeclaude/wakeclaude
English
@dzzgnr love this! now i just need it to buzz me when my session limit resets and we're golden
English
> download the cmux terminal
> immediately impressed
> wonder what all the features do
> can't be bothered to read the docs
> open youtube
> @theo already made a video about it
English
@Hi_Mrinal great solution but one thing i’m curious about is if the db query itself wasn’t optimized what caused the initial request latency to drop from 500ms to 230ms after adding the proxy?
English
Had a chill call with CTO for backend infra role at company in Dublin (January 7th, 2026)
While the conversation got to know they are running ecom dashboards which is taking nearly 500ms for every query (YESS I KNOW YOU KNOW WE ALL KNOW) crashed constantly during peak sales
Built a go proxy with redis on the call while explaining each and every part. The layer resets expiration on each access, the result first request nearly 230 - 249 ms and afterwards it dropped <10ms
it works because the cache learns from behaviour and backend load drops because only misses hit the database.
Cache pattern used ??
> cache aside with time to live reset on access
English
@krushiraj123 my first thought was a cron job too but the thing is cron only works if your mac is already awake. wakeclaude works even when your mac is sleeping because it can wake it up and run the prompt and then go back to sleep again.
English
@_rittik I just made a cron job for the people who want a simple solution this is great IMO
English
fun fact: this project was entirely built with codex
Rittik Basu@_rittik
built a tiny (~1mb) cli tool to schedule prompts for claude code. so now when you hit the rate limit you can just schedule a prompt to auto resume your work. because the work is mysterious & important and must go on. try it: brew install --cask rittikbasu/wakeclaude/wakeclaude
English
also works while your mac is sleeping or the lid is closed. it schedules a wake, runs the prompt, then goes back to sleep.
English
built a tiny (~1mb) cli tool to schedule prompts for claude code. so now when you hit the rate limit you can just schedule a prompt to auto resume your work.
because the work is mysterious & important and must go on.
try it:
brew install --cask rittikbasu/wakeclaude/wakeclaude
English
i recently found out that the govt of india maintains open datasets across various sectors so i thought it'd be a fun side project to pick a dataset and build a chat interface that lets you talk to this data in plain language and get answers.
check it out askmandi.vercel.app
English