Ilir Gruda

Software Engineer.

[Ilir Gruda @ilirvg]

Trying to gain new skills, learn AI, switch careers, or just learn for fun?

Draft and Arc is built to help you achieve those goals.

[Draft and Arc @draftandarc]

ChatGPT is great at answering questions. It's terrible at teaching you something over weeks.

Draft & Arc builds you a full personalized course on anything:
• Notes & highlights
• Study-time tracking
• AI voice tutor
• Feynman-style feedback

Try https://t.co/aisYizT73W

[Ilir Gruda @ilirvg]

@tiwarisuhani_11 This is so cool, you should also check draft and arc
https://t.co/wpnd4QiFRj

[Ilir Gruda @ilirvg]

@tiwarisuhani_11 I really think you should check draft and arc
https://t.co/wpnd4QiFRj

[Ilir Gruda @ilirvg]

@AmjadKheiria you should check this
https://t.co/wpnd4QiFRj

[Ilir Gruda @ilirvg]

@rxhit05 Use draftandarc to master new skills

[Ilir Gruda @ilirvg]

@Alennushkaa Maybe check draftandarc. It is an awesome platform that simplifies the learning process

[Ilir Gruda @ilirvg]

@mert They replace architecture with 'vibes' and wonder why adoption is slow.

[Ilir Gruda @ilirvg]

@NathanGeckler Congrats on the launch! I haven't tried Anything yet.

I'm curious about the 'vibe coding' workflow: Did you have to architect the data structure/state management first, or did the AI handle the structural logic too?

Trying to gauge how autonomous these new tools really are

[Ilir Gruda @ilirvg]

You Architect -> AI Codes
(You define the interfaces/data structures, AI handles the boilerplate)

AI Architects -> You Code
(AI suggests the system design/patterns, you implement the critical logic)

If you let the AI do both you're just generating legacy code at light speed

[Ilir Gruda @ilirvg]

Vibe coding is a trap.

To actually ship production software with AI, you need to split the "Thinking" from the "Typing."

There are only 2 winning workflows

[Ilir Gruda @ilirvg]

@iamarkdev @hytopia What would some of the benefits of using QUIC compared to WebRTC be?

[Ilir Gruda @ilirvg]

@TokunagaKohei Really cool

[Ilir Gruda @ilirvg]

Scaling multiplayer game servers is a different beast

[Ilir Gruda @ilirvg]

We spent months building a multiplayer backend in @nodejs , only to hit a wall at 20 CCU
We debugged, optimized, and profiled... but the architecture just couldn't handle the load.
So we took a risk: rewriting the stack 🧵👇

[Ilir Gruda @ilirvg]

@nodejs @webrtc @nodeclinic @golang @_pion It wasn't that Node is "bad." Node is amazing for I/O-bound web apps. But for CPU-bound, high-frequency packet replication? The single threaded model fights you.

Go’s concurrency model + Pion was the right tool for this specific engineering problem.

[Ilir Gruda @ilirvg]

We spent months building a multiplayer backend in @nodejs , only to hit a wall at 20 CCU
We debugged, optimized, and profiled... but the architecture just couldn't handle the load.
So we took a risk: rewriting the stack 🧵👇

[Ilir Gruda @ilirvg]

@nodejs @webrtc @nodeclinic @golang @_pion It was a 5 day sprint of ~18 hour shifts. But the results were instant.

Node.js: ~20 CCU (Crashes/Lag) Golang: 200 CCU per server

We handled the N² fan-out easily because Go spread the load across all CPU cores.