rust freak |vibing https://nitter.gq/t.co/abpEKRjRCq | One piece addict 💀 | NIT NGP CS'28
India
Joined February 2025
- Tweets4K
- Following730
- Followers549
- Likes9.7K
Pinned Tweet
f*ck it
I am building clash royale in rust form scratch
currently building the game env
using bevy
will post everyday my progress and journey
documenting the whole precess in md files
Parth Agrawal retweeted
vidya-bodha.com is live.
My GitHub repo crossed 830+ stars, so I turned it into a proper learning website.
70+ in-depth engineering articles, all in one place, organized so you can actually learn instead of scrolling a timeline.
What you’ll find:
-30+ HLD breakdowns (YouTube, Netflix, Spotify, Instagram, Amazon, Discord, and more) -LLD classics like parking lots and vending machines -Backend fundamentals: APIs, databases, caching, concurrency -A complete Distributed Systems series, all 9 parts -A complete Microservices series -Search and filters by topic -Structured learning paths -Progress tracking, so you know what’s done and what’s next
“Vidya” means knowledge and “bodha” means understanding. That’s the goal here: not just reading about systems, but really understanding how they work.
AI, ML, and DevOps content is coming next, and every new article lands on the site first.
The GitHub repo stays open if you’d like to star it
github.com/harshit3011/Techn…
Parth Agrawal retweeted
I'm begging you to try systems programming.
I built tiny AWS - a cloud using no Docker, no VMs, just 10,000 lines of Golang and Rust code with this same mindset. Tiny AWS is legit just cli, and not even React or NextJS codes.
Stop spending all your time learning another framework, another library, another abstraction. Go figure out what actually happens underneath.
- Write your own memory allocator.
- Build an operating system kernel.
- Implement a TCP stack.
- Write a compiler.
- Build a database engine.
- Understand how Linux schedules processes, manages memory, handles interrupts, and talks to hardware.
Learn C. Get your hands dirty with Rust. Read source code. Read the Linux kernel. Break things. Spend time debugging a segmentation fault because you forgot how pointers work.
And then experience the satisfaction of finally understanding WHY something works, instead of just knowing WHICH API to call.
We're entering an era where AI can generate thousands of lines of code in seconds. Frameworks are getting easier. Abstractions are getting higher. The barrier to shipping software is collapsing.
But here's the interesting part:
The easier it becomes to generate code, the more valuable it becomes to understand what's happening underneath it.
Anyone can ask an AI to build an application.
But can you explain WHt happens when a process gets created? How memory is allocated? What happens when a packet travels through a network? Why a database quEry becomes slow?
Can you debug something when the abstraction leaks and the documentation stops helping?
That's the difference between knowing how to use technology and understanding how to engineer it.
You don't need to become a kernel developer. You don't need to abandon web development. And you certainly don't need to reinvent everything.
Just spend some time building things that force you to think.
Because there's something incredibly beautiful about programming when you stop treating computers as magical boxes that execute your JavaScript.
Don't just learn to build software. Learn how the machine works.
I'm telling you, try systems programming at least once.
Finally all quiz and evalvs done today 😭
Now just endsems papers are remaining all this shit ate a lot of my time
But today I am going home 🥹
☠️☠️
a friend of mine from IIT BHU told me about recent Flipkart AI engineer OA
And guess what it had DSA round
And even in that, one question was rated 2500+ in codeforces
So basically, companies now want entire AI engineer stack ( modelling, pipelines, evals, testing and scaling) and also want you to be above Master in CF
Is this a joke?
Parth Agrawal retweeted
Today we’re releasing Sarvam Prism, a System One model built on Sarvam 105B for fast, accurate decisions across enterprise workflows.
It scores 71.8 with a median response time of 44 milliseconds, the highest average score among decision models that respond in under 100 milliseconds.
Bhai bas khatam karke jana 😭
#ONEPIECE1198
Seems like the new schedule is:
Two chapters+Break
Before it was 3 Chapters+Break.
It’s a bit concerning, honestly.
The number keeps dropping every year; 10 years ago, it was 5 chapters followed by a break...
You can tell that Oda is getting older.
Me after unlimited tokens
Weekend hack: HackerTube. I was annoyed with YouTube's recommendations so I had Claude filter down 43M comments over 20 years of HN to 6,454 videos that keep being recommended. *Significantly* better viewing experience, ~2 hours with Opus 5.5:
hackertube.org/
- Excellent recs without any logins (state is kept in local browser session)
- Thumbnail is a representative frame from middle of the video instead of the super duper A/B test hyper-optimized user uploads
- Underlying data from the comment analysis as one zip file, point your Claude at it and ask for your own version
- HackerTube Longs ✊
A bit more on the process:
I use Claude Code (desktop app for Mac), Opus 5.5 on Medium effort as my default. I only talk to Claude and have it do everything, the only "manual work" I helped with in tandem was in the browser (reminiscent of MenuGen): 1) I had to create some kind of API key from Google Cloud UI to query YouTube for video metadata, 2) I booted up a VPS box for serving and 3) I bought the domain name. Once the box was up I gave Claude the IP and told it update, secure, harden, deploy and optimize the code, and it did the whole thing.
Many awesome projects that would have normally taken weeks/months even can now be few-shotted for breakfast. What else are we not thinking of.
i am thinking to make a ai agent which handles my whatsapp and handles all these college groups regarding evals and assignment solutions
Parth Agrawal retweeted
Someone turned the hardest parts of LLM inference engineering into a course where you have to fix the performance problems yourself.
And you can do it on a Mac.
The project is called tiny-llm, an open-source course for building a miniature vLLM-style inference engine around Qwen3.
It starts with a simple implementation using matrix operations. You build attention, normalization, positional encoding, the Transformer layers, and the loop that generates tokens.
You have a working model.
Now the interesting engineering begins.
During decoding, repeatedly processing the previous tokens wastes computation. So you implement a KV cache and reuse previously computed attention state.
Next, you look at model weights. Keeping everything in dense BF16 consumes memory bandwidth. So you retain packed 4-bit weights and implement specialized matrix operations.
Then you move into Metal kernels, SIMD matrix operations, fused RMSNorm, RoPE, SwiGLU, and tiled prefill attention.
Once the single-request engine works, you tackle a different problem.
What happens when multiple requests arrive, each with different prompt lengths and different amounts of cached state?
You implement continuous batching, chunked prefill, paged KV storage, and attention that reads directly from noncontiguous cache pages.
The four-week curriculum eventually extends into a small coding-agent harness with validated tool calls, checkpoints, context management, and evaluation.
Every stage comes with exercises, reference implementations, tests, and measurements to help you understand the behavior of the system you're building.
The course uses MLX, C++, and Metal on Apple Silicon. You don't need an NVIDIA GPU, and the introductory path supports Qwen3-0.6B on an 8GB Mac.
Some later checkpoints remain under development and review, but there's already a substantial amount of real inference engineering to learn from.
The GitHub repo has crossed 4,800 stars.
GitHub: github.com/skyzh/tiny-llm
