aadiii (@microtaskq) header
aadiii (@microtaskq) profile picture

emotionally attached to silicon based lifeforms swe @ https://teapawt.org/t.co/Ts9B286tax pfp by @chocobidealer

undefined
Joined December 2023
Photo by aadiii (@microtaskq)Photo by aadiii (@microtaskq)Photo by aadiii (@microtaskq)Photo by aadiii (@microtaskq)Photo by aadiii (@microtaskq)Photo by aadiii (@microtaskq)Photo by aadiii (@microtaskq)Photo by aadiii (@microtaskq)Photo by aadiii (@microtaskq)Photo by aadiii (@microtaskq)
aadiii retweeted
Hey everyone, I am currently looking for my next role in backend, systems, or ai infrastructure engineering. preferably remote-first. i'm based in india (ist) and comfortable working across us/eu time zones. i mostly work with go, along with python and typescript. i enjoy building things from scratch, understanding how they work under the hood, and occasionally going a little too deep into the details. some things i've built: mini-rpc — an rpc framework in go over raw tcp, with custom framing, serialization, and code generation. mini-protoc — a lightweight protocol buffers compiler in go for mini-rpc, with a hand-written lexer, parser, ast, schema validator, and go code generation for message structs and service/client stubs. pony — a go runtime for orchestrating multiple ai coding agents, with process lifecycle management and pty support. krafta — a kafka-inspired log system in java, with partitions, offsets, consumer groups, and multi-broker coordination. newsletter backend — built using go, postgresql, rabbitmq, and redis, with a focus on async processing, idempotency, and rate limiting. recently, i've been spending more time exploring the infrastructure side of ai —agent runtimes, llm gateways, mcp, and the engineering challenges involved in making these systems reliable. i'm open to junior / entry level opportunities in backend, go, systems, or ai infrastructure. remote roles would be ideal, but i'm also open to opportunities in bengaluru/hyderabad. if you're building something interesting, know of an opening, or think there might be a good fit somewhere, feel free to reach out. referrals and introductions are always appreciated. website: akansharoy.site
github: github.com/akansha204 my dms are open!
55
46
5
354
14,948
非常开心地分享我与 @MeshyAI 团队合作的最新技术成果,兼容 three.js 的实时光追引擎 ThreeRT! 基于 ThreeRT,哪怕是非常简单的 three.js 场景代码,都能获得稳定的高质量 3A 级光追效果,而不必了解全局光照、阴影贴图、反射探针等传统 three.js 光栅化调教底层细节。
What if three.js could render beyond the limits of the browser? Meet ThreeRT: a native JS runtime and hardware-accelerated real-time path tracer, bringing lifelike lighting to Three.js scenes on macOS, Windows, iOS, Android, and Steam Deck. Fully open source 🧵
46
57
16
690
45,601
aadiii retweeted
What if three.js could render beyond the limits of the browser? Meet ThreeRT: a native JS runtime and hardware-accelerated real-time path tracer, bringing lifelike lighting to Three.js scenes on macOS, Windows, iOS, Android, and Steam Deck. Fully open source 🧵
37
139
35
1,505
110,045
ive been working on a full architectural visualization for the mi300x card. this covers everything you need to visually understand about the architecture starting from individual lanesthrough chiplets, compute units, wavefronts, matrix instructions, and the memory hierarchy. you can rotate and explode the model, inspect components, experiment with execution masks, see how matrix elements map to lanes, and calculate LDS usage for different tile sizes. there’s also an animated trace following a tile through memory to matrix computation, plus comparisons with H100 to connect familiar NVIDIA concepts to AMD’s architecture. project link in the comments -
2
5
1
67
1,713
a very gentle introduction to AMD GPU programming and transition from CUDA kernels
10
28
1
197
5,288
im building noro - a image and video inference library. it is built on top of huggingface diffusers with a focus on performance. i currently have the krea-2 model implemented and across all image dimensions, on average we have a 20% speedup. im currently building it for nvidia chips and also extending to amd as well. all the kernels and tricks are written in triton and gluon! repo link in the comments.
6
6
1
123
6,646
aadiii retweeted
Introducing Lycan !! [Link in replies] A Game Boy Advance emulator built in Rust for Linux native and Browser You can play games straight from your browser, just upload a BIOS file and a GBA game file Link to Browser version and Github in the replies You can download the Linux native file and run it locally too (instructions in Readme and releases)
25
18
3
104
4,922
why is it ..
2
1
20
264
aadiii retweeted
hi skribbl (made in rust btw)
29
6
188
6,930
Wallets, in raw CSS. Gradients are very underrated. If you layer enough of them you can mimic depth and texture in 2d space
6
3
38
1,387
i hit cuBLAS level performance on Ada Lovelace using CuTe DSLs kernels in python i walk through a series of 6 kernels, show how to profile them and understand profiling bottlenecks, tons of visuals and how to write understand CuTe DSL kernels in python. i show that you dont need to write kernels in C++ to get high performance but you can do it in python. blog - jino-rohit.github.io/blogs/0…
15
27
5
214
16,611
all good
1
5
1
28
2,982
aadiii retweeted
Introducing kinctx kinctx is a terminal-first command library for the snippets you keep rewriting, re-Googling, or forgetting. It let's you save snippets and edit placeholders before executing them check it out here: kinctx.kaizakin.site/ Github link in the comments
25
11
4
123
22,162
aadiii retweeted
1
10
949
aadiii retweeted
lil tease of what i've been building over the past few weeks: starling. adk for go projects but what's different is that i've taken a page from temporal's book around durable execution and event history: durable logs, replayable runs, tool use, budgets, and if it crashes mid-run you can resume from the last recorded event. made for simple agentic use cases that still need proper replayability and auditability, so there's always an answer to "why did the agent do that here?" and also when. there's still a lot to test and flesh out. the way to do that is to build something else with it, which i'm also working on btw. have asked some friends trying it in their projects too, which should make my life easier. plan to spend the next couple of weeks fleshing this thing out. the amount i've learnt building it is insane and i still got so much to figure out. more about it soon but yeah.
10
12
68
2,772
aadiii retweeted
Introducing Dotmatrix🗿 A collection of 55+ free and open-source dot-matrix loaders, built with React, TypeScript, Tailwind CSS, and shadcn. Install one, copy the code, and make it yours. Link: 👇🏼
91
134
14
2,392
148,086
Built A-WAL an append only write ahead log in Rust [Link in replies] i built A-WAL with one thing in mind: it should be a trustworthy byte log that can recover a database from crashes, repair torn tails, replay history, and prune old segments safely these are the features i implemented in the wal: - frozen segment and record formats - monotonic LSN allocation - append and batch append - explicit flush() and sync() durability boundaries - segment rollover with segment seal records - startup recovery scan - corrupt tail truncation - corruption detection in sealed history - sequential replay - point lookup by record LSN - clean shutdown witness and fast reopen path, - whole segment retention pruning, - retention pins for long-lived readers - concurrent access through WalHandle - tail following iteration - deterministic crash/fault injection - metrics and recovery reporting I have tried to cover all this in detail inside the project Readme, do check it out ;) Benchmark results are also pretty good :)
9
20
2
186
10,444