Angshu (@angshuhere) header
Angshu (@angshuhere) profile picture

THE LAST DANCE

subjective reality
Joined November 2022
Angshu retweeted
introducing kibu, a free and open source SI desktop pet that does real work on your mac. → press ⌘⇧space and just ask → uses jev for fast decision making → plans your day around your calendar → remembers your preferences, gets smarter over time → finds, renames and sorts your files, with undo → never touches your mouse → bring your own agent: claude code, codex or opencode → it's everything in spotlight + more (just as fast) → your history stays on your mac → uses <1% cpu when idle getkibu.vercel.app
30
8
5
93
7,985
Angshu retweeted
Two execution traces can look completely different even when nothing important changed. Timestamps, random IDs, and event ordering create noise that makes real changes hard to spot. Built TraceDiff: feed it two traces, get back only what actually changed. Demo Link Below
12
24
3
42
3,224
Angshu retweeted
Releases tomorrow.
5
1
1
45
1,305
Angshu retweeted
Exploring visual identity for the first time rate this out of 10
50
30
1
552
11,894
Angshu retweeted
made a little guy. he's cooking something. i'm not allowed to say what.
7
2
1
44
1,617
Angshu retweeted
TraceDiff Chapter 1: The Problem Say you're working on a repo and you refactor a variable name. You run a diff, and it shows two lines changed. But you're looking at that diff wondering: is this actually something I need to care about? 🧵 1/7
5
21
1
32
1,338
Angshu retweeted
i'm working on kivo, an in-memory kv store and till now we're here - min-heap for TTL with lazy checks on read ops like get/exists and active removal through a bg routine - bucket striping to split data internally and improve concurrency - per-key write path handling new inserts, updates and TTL transitions (permanent -> expiry, expiry -> permanent) while keeping size and heap state consistent - random sampled LFU eviction with background decay to keep memory bounded it's still wip and more updates will come
2
3
18
326
Guys, I finally published my first video on YouTube & Instagram. 🥹❤️ I’ve never really asked you guys for anything, but this time I genuinely want your love & support. Art & craft has always been a part of my life. I’ve always wanted to do something with it, but never really got the opportunity. Being the elder daughter, I was mostly surrounded by the responsibility of studying well, getting a stable career, and being “practical.” But this time, I thought… why not give my dream a try? 🥹✨ So here’s my first little step. 🎨🌸 If you can, please watch, like, share, or even just leave a comment. It would genuinely mean a LOT to me. ❤️ YouTube: youtube.com/shorts/DaXtvv1Q6… Instagram: instagram.com/reel/DdB4vMhTs…
51
15
3
112
4,373
new library at our college is actually so peaceful thankfully no couples yet
14
1
119
2,641
Angshu retweeted
An engineer mostly does three things: read, build, and write. i've always liked all three: - reading about systems and ideas - building things to understand how they actually work - writing about whatever i figure out along the way so i made a little place around that. stuff i build, things i read, things i write, along with some of my projects and work. finally putting it out there, my portfolio is finally live : check the link in the thread.
37
34
6
79
9,702
I'm currently looking for opportunities to work on and build interesting things. Here's my portfolio: lowkeydev.me/ It has everything about my work, projects, and experience. If you think I could be a good fit for your team, feel free to DM me. Would love to chat!
10
17
2
56
3,654
There is still so much more, more life left to live, more love to be felt, more ppl to meet, more doors to open.
2
28
2,504
got way too many things going on rn, my energy is spread across everything and I am finishing nothing lol
5
32
710
OK I drew a rough high level workflow diagram by hand so everyone can understand what the project is about. so basically, it's kind of a banking/ledger system where we can deposit, withdraw and transfer money. right now as you can see in the diagram, we have one leader and two followers. the leader accepts client requests, let's say on port 9000 and listens for follower connections on 9001. The two followers expose their own client ports, 9010 and 9011. when a client makes a write, let's say a deposit, the leader first writes the event to its WAL on disk and updates its in-memory state. After that, it broadcasts the event to the followers and they store the replicated event in their own WAL and apply it to their own in-memory state. we can read and write from the leader, but the followers are read-only. So all writes go through the leader. so it's a a single leader replication setup. And when we restart a node, it reads its WAL and replays the events to rebuild the in-memory state(event sourcing). So even though the actual account balances are in RAM, we don't lose them after a restart because we can reconstruct them from the WAL. I haven't implemented leader failover yet. We can currently assign a new leader manually but I am working on implementing automatic leader election, where one of the followers becomes the new leader after the current leader fails.
Knot progress: So last month in august I made the mvp of this project (still need to improve a lot of things), but everything was in memory. For the past couple of days, I have been working on implementing WAL. Previously, the log and state lived only in memory, so whenever we restarted a node, it lost everything. Now added a basic WAL storage layer. Whenever a new LogEntry is committed, it gets serialized and appended to wal.log using length prefixed records. The data is flushed and synced before the write is considered successful :)
1
46
1,380
u can also try it by urself u should see something like this:
4
137
editing 1000 images at the same time in a single canvas @OffToning
34
22
4
121
6,986
Knot progress: So last month in august I made the mvp of this project (still need to improve a lot of things), but everything was in memory. For the past couple of days, I have been working on implementing WAL. Previously, the log and state lived only in memory, so whenever we restarted a node, it lost everything. Now added a basic WAL storage layer. Whenever a new LogEntry is committed, it gets serialized and appended to wal.log using length prefixed records. The data is flushed and synced before the write is considered successful :)
4
3
1
39
2,719