UK landlords: first Making Tax Digital quarterly update is due Friday 7 Aug.
The bit that catches people out - figures are cumulative from 6 April, not just the last 3 months.
No penalty points in year one though.
Spreadsheet I built for it: https://t.co/zE60a9fk36
@danielkleach all four boxes checked. building Nimbus, a chat UI kit for AI apps, desktop-first because every competitor mock is mobile. lite version free on GitHub, full kit on Gumroad. traction status: still introducing myself to strangers on the internet
small account observation after a few weeks: my posts barely leave the building, but a decent technical reply in someone else's thread actually travels. the algorithm pays you for joining conversations, not starting them. adjusting accordingly #buildinpublic
@PeilvDog_CN hit the same wall with a scan daemon. one slow model response inline and the whole loop stalls past its window. moved anything slow into spawned child processes with their own timeouts, parent just schedules and moves on. also means code edits get picked up fresh per child
@Subhrodeep1905 system design, easily. I run a few small agents 24/7 and every failure that hurt was plumbing, not prompts. bad state handling, retries with no guardrails, a daemon reading stale config. prompt tweaks moved quality maybe 10%. boring architecture keeps it alive overnight
for anyone who's built an AI chat frontend: what's the one feature you always end up re-adding from scratch? for me it's the "stop generating" button. forget it every single time. #buildinpublic
@adixranjan08 building it yourself teaches way more than wiring up a lib. the bit that caught me out: rendering markdown while tokens are still streaming. re-parse the whole message on every chunk and it flickers + tanks the framerate. worth solving early rather than at the end
@rahilpirani@SecondBrainCF not sure it has a name, but the usual cause is the daily digest email — it goes out the morning after and reaches a lot more people than the site does on the day. the newsletters that scrape the leaderboard land then too. congrats on the top 3 btw
@jourverse pipeline first is the right order imo. fwiw three already frustum-culls per object — the thing that quietly kills that is merging everything into one mesh, since you end up with one giant bounding sphere and nothing ever culls. worth checking before hand-rolling it
@RahulSain714 the axis nobody puts on these charts: how much of it you can hold in your head at 2am when it's broken. solo or small team, that one beats scalability every time. you can split a monolith later, you can't un-split 12 services in a hurry
@Folkodeur which hours actually produced something i shipped, vs which ones just felt busy. every tracker i've used tells me where the time went but never which of it was worth anything. that gap is the whole product imo
the trick to making an AI chat feel "live": don't wait for the full response. read the fetch body as a stream, decode each chunk, append tokens as they arrive. ~10 lines of vanilla js, no library. that's the whole magic behind the streaming caret.
the trick to making an AI chat feel "live": don't wait for the full response. read the fetch body as a stream, decode each chunk, append tokens as they arrive. ~10 lines of vanilla js, no library. that's the whole magic behind the streaming caret 👇 #buildinpublic
live demo, no signup: https://t.co/ulkzREfNyW
code: https://t.co/KyUCIobdFQ
there's a paid full kit too (citations, reasoning blocks, tool-call chips, attachments) but the free one stands on its own 🙂 #buildinpublic#webdev
i kept rebuilding the same AI chat UI every project. streaming, code blocks, dark/light toggle, all from scratch 😮💨
so i built one clean version you just drop in. plain HTML/CSS/JS, no framework, no build step.
free version on github 👇
I kept rebuilding the same AI chat UI for every project, so I made one clean drop-in kit.
Streaming, reasoning + tool-call blocks, code blocks, light/dark + 5 themes. Vanilla HTML/CSS/JS — no framework, no build.
Just shipped it 👇
https://t.co/o3mCcytmcD
#buildinpublic