SD round at Meta for L5 AI/ML (CTC: 1Cr+):
Meta shows you a reel from an account you never followed, never searched, and have never interacted with. You watch it 3 times in a row.
Meta knew you would before you did.
It did not use your follow list, your search history, or your likes.
What signal did it use, and how do you design a recommendation system that knows what you want from behaviour you never consciously performed?
“Cut every engineer who is not at 100% AI output” is exactly how companies end up with code nobody understands and outages that take days to fix.
It’s funny because these hot takes blindly miss the point.
A software engineer does not do one kind of work all day.
Sometimes they write boilerplate.
Sometimes they write tests, refactor old code, generate docs.
AI can absolutely help here, do it even 10-20x faster.
But sometimes they are designing a payment flow, debugging a production incident, or figuring out why a small edge case can corrupt data for thousands of users.
That is not the same work.
The impact is largely different.
That's why teams should not measure engineers by how much code AI wrote for them.
They should measure whether engineers used AI in the right place, understood the output, and owned the consequences after shipping.
I would even argue that, an engineer avoiding AI completely is a problem. But, an engineer trusting AI blindly is a bigger problem.
Because AI can sound confident while making the wrong abstraction. It can generate 600 lines for a 60-line problem.
It can miss business context, ignore old system behavior, and create bugs that only show up under real traffic.
And then someone still has to debug it.
Someone still has to explain it.
Someone still has to take responsibility when customers are blocked.
I use AI heavily, but I don't surrender my judgment to it.
As an engineer, you need to know when to ask AI for speed and when to slow down and think. It's only huge consequences if you don't draw that boundary.
Many high-performing professionals get promoted into leadership roles every year, but mentally, they still operate with an individual contributor or employee mindset and not a leader mindset.
This promotion doesn’t teach you leadership overnight. You weren’t formally trained to handle the hardships, challenges, and pressure that come with the role.
That’s why leadership starts feeling overwhelming.
You keep:
• Doing everything yourself
• Avoiding difficult conversations
• Overthinking every decision
• Constantly seeking validation
• Carrying your team’s pressure emotionally
Because leadership is no longer just about execution.
It’s about giving clear direction to your team, taking ownership of the entire team’s KRAs, making tough decisions, staying emotionally balanced in difficult conversations, showcasing performance in front of senior leaders, and leading people under pressure.
Leadership notices this shift very quickly.
At senior levels, people aren’t only evaluating your performance.
They’re evaluating how you think and operate as a leader.
If you’ve recently stepped into leadership but still feel stuck in execution mode, overwhelmed managing people, or unsure how leaders actually operate — it’s time to build leadership readiness, not just performance.
System design is a lot like solving a puzzle 🧩
Your building blocks are the pieces scattered on the table. Design patterns are the guide that shows you how everything actually fits together.
You are in an MLE interview at Perplexity and the interviewer asks:
"Our RAG system pulls the right documents every single time. The correct answer is sitting right there in the context. The model still gets it wrong."
You say: "The LLM is hallucinating. Better prompt or better model."
Interviewer says they already tried both. Still wrong.
You are now out of answers.
What is actually breaking between the retrieved document and the final response?
Interviewer at Meta for E5 (Sr. Engineer)
WhatsApp says "You may have new messages."
Not "You have 3 new messages." Not "You have no messages." Just... maybe.
WhatsApp delivers messages to 2 billion users in real time with perfect tick accuracy.
How does a system that tracks every single tick across 6 servers not know whether your messages actually arrived?
Btw, if you’re preparing for Senior to Principal-level system design interviews, I’ve put together 90+ fundamentals like this into a guide or meet with me for high quality mock interviews/mentorship.
You can check the details here: https://t.co/1JIogXSDjf
Without data compression, the modern internet completely breaks.
~ YouTube couldn't serve billions of streams.
~ Spotify couldn't host millions of songs.
~Your phone couldn't hold thousands of photos.
~ The web would load 10x slower.
Compression is the invisible backbone of systems engineering. Without it, none of our current cloud infrastructure works.
Why should a systems engineer care?
Because every byte you send over the wire carries a latency and financial cost. Understanding how data behaves allows you to choose the right serialization and compression strategies for your payload.
Next time you optimize an asset pipeline or an API response, look closely at your data density. The math saving your cloud budget is beautiful.