Abhishek Dhiman

🧑‍💻Software Engineer & Community Manager | Ex @accenture | Ex TA @CodingBlocksIn | Member of Google Developer Group Noida & Delhi

[Abhishek Dhiman @codedhiman]

Watched a video on how LLMs actually work. Here what I Learned:

Text gets split into tokens, each token becomes a vector (embedding + positional encoding so it knows word order).

From that, every token gets a Query, Key, Value vector. Q x K gives attention scores, how much each token should attend to earlier ones (masked, no peeking ahead). Add the input back in, layer norm, feed-forward, add+norm again. Repeat across dozens of layers.

Also TIL GPT-style models are decoder-only, no encoder. The original "Attention Is All You Need" paper (Google, 2017) used encoder-decoder for translation, decoder-only was enough for generation.
At the end it just predicts the next token, feeds it back in, predicts again. Repeat.

Killed a few assumptions for me: no real memory (it just re-reads the whole convo each time), bigger model isn't "trying harder" it just saw more patterns, and there's no actual understanding in there. Just very well dressed-up prediction.

[Abhishek Dhiman @codedhiman]

12/12

If your company is hiring for a Software Engineer or Backend Engineer role, I'd really appreciate any referral or lead.

A simple DM could make all the difference.

Thanks for reading, and I hope next time I post, it's about finally getting that offer. 🚀

[Abhishek Dhiman @codedhiman]

1/12
12 months. Still no offer.

I left my job in August 2025 to upskill and look for better backend engineering opportunities.

I thought it would take a few months.

It's been almost a year.

[Abhishek Dhiman @codedhiman]

11/12

If you're also job hunting right now:

Keep learning.

Keep building.

Keep practicing.

Especially DSA if you're targeting product companies.

One interview can change everything.

[Abhishek Dhiman @codedhiman]

@Arpansac @Samaytwt True!!

[Abhishek Dhiman @codedhiman]

@satakshi_ai AI h

[Nandkishor @devops_nk]

Best visual to understand a CI/CD pipeline for begainers.

- Source Code (Git commit / PR)
- Code Review & Merge
- Build (compile, package artifact)
- Static Code Analysis (SAST, linting)
- Unit Tests
- Integration Tests
- Artifact Storage (Docker registry / artifact repo)
- Security Scans (SCA, container scanning)
- Deploy to Lower Environment (Dev / QA)
- Automated Testing (E2E / regression)
- Approval Gate (manual / automated)
- Deploy to Production (rolling / blue-green / canary)
- Monitoring & Alerts (logs, metrics)
- Rollback Strategy (if failure)

From commit → tested → secured → deployed → monitored.

That’s a real-world CI/CD pipeline.

[Abhishek Dhiman @codedhiman]

I'm thinking of making a small, fun project. Which AI vibe coding tool should I use? #AI

[Abhishek Dhiman @codedhiman]

@ColossalMonk 👾