Nischit Shetty

SSE-I at Geekyants | React | React Native | Ex-Amazon

[Nischit Shetty @nischitcodes]

@NammaBESCOM @16 This is now resolved. Thank you @NammaBESCOM

[Nischit Shetty @nischitcodes]

Powercut in BTM 2nd Stage near BTM library since more than 1 hour. Please look into this @NammaBESCOM

[Nischit Shetty @nischitcodes]

@NammaBESCOM @16 The power is back but now low voltage issue

[Nischit Shetty @nischitcodes]

@NammaBESCOM no power in BTM 2nd Stage mico layout. Please check

[Nischit Shetty @nischitcodes]

Hey @GuiveAssadi,
Not applying for the role, but the @MechanizeWork take-home looks like an awesome exercise. Any chance I could run through it just for fun/practice?

[Nischit Shetty @nischitcodes]

@Ctrlmemes_ I am a little bit surprised that you missed the point where officials went bonkers over a joke. So many crime in the country and all the officials went bonkers over a comedian
No wonder the country is in ruins when there are people like you

[Nischit Shetty @nischitcodes]

This week was about building the mental model: Transformers → Tokens → API → Streaming → Local models.
To follow my journey check this blog: Documenting My Journey of Learning AI (https://t.co/Imjr7CP8zd)

[Nischit Shetty @nischitcodes]

I spent years as a Senior Developer, but I realized I was just "using" AI without knowing how it actually worked.

So, I went back to basics to rebuild my mental model from scratch.

From Transformers to Token IDs here’s what I learned : 🧵

[Nischit Shetty @nischitcodes]

The biggest game changer? Running models like Llama 3 locally via @Ollama.
- No API costs
- Zero data leaves your machine
- Standard OpenAI-compatible endpoints
- Perfect for prototyping and privacy-sensitive dev work.
6/7

[Nischit Shetty @nischitcodes]

As a dev, this hit home. Waiting 8s for a response is bad UX.

Streaming sends tokens as they’re generated. That "typing" effect isn't just for show—it makes the AI feel alive and cuts perceived latency to zero. It's the standard. 5/7

[Nischit Shetty @nischitcodes]

Computers don't see "Apple." They see a Token ID (a unique integer).

Every model has a "vocabulary" lookup table. The ID for "Apple" in GPT-4o is different from Gemini. To the AI, your blog is just a long list of integers. 4/7

[Nischit Shetty @nischitcodes]

I used to think 1 word = 1 token. Nope.

Tokens are chunks of characters. "Learning" might be one, but "unbelievable" could be three.

This matters because API costs and context limits are token-based, not word-based. 3/7

[Nischit Shetty @nischitcodes]

Neural networks were great, but the real shift happened with Transformers.

The "aha" moment? They don't read word-by-word. They use "Attention" to look at the entire sentence at once for context. This parallel processing is why AI scaled so fast. 2/7

[Nischit Shetty @nischitcodes]

@CinemaXCricket It is funny how we will keep doing this to comedians but never to the local MLAs, MPs or ministers. Even street dogs have less malice than you

[Nischit Shetty @nischitcodes]

@APamnani @SatanAtWink @grok can you tell us about the road accidents in UAE and compare it with the numbers in India?

[Nischit Shetty @nischitcodes]

@sagorika_s

[Nischit Shetty @nischitcodes]

@kissingerspeaks This is racism at peak. There are a lot of scammers out there - Mozaic Payments, FTX Crypto scam but none of them were targeted for their origin or race. The moment it is Indians it is okay to do racist comments. And among them are crabs like you!

[Nischit Shetty @nischitcodes]

@NammaBESCOM No progress yet @NammaBESCOM

[Nischit Shetty @nischitcodes]

@NammaBESCOM it has been more than 5 hours since power cut in 13th A cross road. 1912 isnt working neither are any Bescom folks responding. Please check

[Nischit Shetty @nischitcodes]

@NammaBESCOM Shared details via DM. Please checm

[Nischit Shetty @nischitcodes]

13th A cross road, BTM 2nd stage, Mico layout