Vishal Sahoo

WILL BE CONNECTING TO THE PEOPLE WHO ARE IN TECH. ONLY TECH !!

@salujamehak5 Who gave the name automata

very easy 🤧

@salujamehak5 mcAfee is the proof you aren't doing enough development too in ur laptop.

zindagi mushquil hai!!

@salujamehak5 Worst part of college

wtf is this 🥀😥

@GurugramDeals @Benarasiyaa IG you are the owner of the car

@Benarasiyaa His fault, you can't be standing in the middle of an expressway!!

@GurugramDeals @Benarasiyaa Wow man donate your eyes

@Saswat_Mishra22 🤡

@Unowmeno_7 @grok Saswat Mishra 🤓

Hey @grok who is the most famous person who visited my profile, don't tag them , just give me a name!?

Hey @grok , hows current tech market? Tcs to layoff 12k people, microsoft to 20 k people, so do intel...? What the situation, trumph advices not to hire indian talent..whats the % of an avg indian youth getting a job in cse or engineering ? Hows the fight for govt jobs in India?

Damnn!! @Replit

This was my last project 😂🥲

Don't knw why but late night coding sessions >>> netflix chill Not every weekend requires netflix and chill

RoPE (Rotary Positional Embeddings) is a technique that adds relative position info directly into attention scores using rotation math Helps models understand sequence order Scales better for longer contxt lens Replaces older sinusoidal/abslute embeddings LLMs cn rembr 100+tokn

@sreeeja_ so what are you currently learning

@realsohamparekh Clean hacks fade, solid systems last.

over-engineering is underrated in this world full of vibe coding

@realsohamparekh very true :(

Having a lean team (of cracked people) is a competitive advantage

MoE (Mixture of Experts) blew my mind. Instead of using the entire model for every input, it activates only a few expert paths. You get massive models with less compute. Faster. Cheaper. Scalable. Feels like the smartest way to scale LLMs right now. #AI #MoE #LLM #DeepLearning

@abhisheknaironx To stay at the top, you always need to upskill Doesn’t matter if you’re in a job or jobless. The grind never stops 🧠⚙️

Hit me with one harsh truth you've accepted so far as a dev.

@AdityaShips This gives me hope

Started with 0 followers on Jan 1, 2025. Barely touched 30 followers in the first 5 months. Then something changed. In the last 45 days — I crossed 5000 followers 🥹 When consistency meets clarity, growth feels like magic.

@sreeeja_ heads up

Training gets the spotlight, but inference is where real users meet your model. You can train a 70B LLM, but can you serve it to thousands with low lat and low cost? That’s where tools like vLLM, TGI, and TensorRT-LLM enters. Smart batching, KV cache reuse, token streaming tops.