Ankit Kumar

You are in a system design interview at Amazon for an SDE-3 role and the interviewer has given you the question to Design Netflix. He then asks a follow up to you: How does Netflix knows when to show: Are you still watching? Is it just time-based? If it were time-based, you would see it every time. So what is actually going on under the hood? Both look like a simple pop-up. Underneath, it is a mix of product thinking, client logic, and backend events. Btw, if you’re preparing for system design/coding interviews, check out our mock interview tool. You can use it for free here: https://lnkd.in/gpCn7t2T [1] Start from the product goals The feature is not only about nagging people. - Save bandwidth and CDN cost if the viewer has slept or walked away - Avoid autoplaying potentially sensitive content in an empty room - Protect kids if parents start a show and leave the TV on - Do all this without annoying active binge watchers So the design must answer one question: "When is the user probably not here any more" [2] Naive design - pure timer Simplest idea: - If playback has been running for 2 hours, show Are you still watching - If user clicks "Yes", reset the timer Why this is weak: - Someone can binge 5 episodes in a row and gets interrupted in the middle of an intense scene - Someone can start a show, walk away after 5 minutes, and the platform will keep streaming for the next 2 hours - It ignores how many episodes were auto played, device type, time of day, user habits  [3] Realistic design - session and engagement signals Think in terms of a "watch session" and "engagement events". Signals the client can track: - Play, pause, seek, volume change - Episode finished and next episode auto started - Remote or keyboard input, UI navigation - Screen on or off events from the device - For mobiles: app background or locked state A common heuristic could be: - Count how many episodes have auto played without any user interaction - Track how long it has been since the last button press or navigation - Only trigger the prompt when both are high enough Example rule: - If 3 episodes in a row have auto played - And there has been no interaction for 45 minutes - Then, before starting episode 4, show Are you still watching This feels much smarter: - Active viewers who pause, skip intro, change volume keep resetting the "engagement clock" - Sleeping viewers do not touch anything, so the next episode is blocked by the prompt  [4] Client heavy vs server heavy design You can talk about two design choices. Client first: - All logic runs inside the app on TV, mobile, or web - Client keeps an in memory session model and decides when to show the popup - It still sends telemetry events to backend for analytics and future tuning Pros: - Works even with flaky network - Highly responsive, no extra server round trip Cons: - Logic must be implemented and updated across many platforms - Harder to quickly roll out rule changes Continued ↓

184

12 Comments

Micron, SK Hynix, and Samsung stocks are soaring but few understand why.   1. The structural reason:   - Memory, in particular High-Bandwith Memory (HBM) has become crucial to run LLMs for billions of users - Running LLMs is mostly is memory-bandwidth bound, not compute-bound - During decode, GPUs spend more time fetching weights and KV cache than doing math, making HBM the primary bottleneck   2. The supply chain reason: - As demand soared, the major players shifted production capacity to high-margin HBM - That led to undersupply of other memory types (SRAM, DRAM) which are still needed for AI   What this means for…   1. Business leaders Memory cost will drive up GPU pricing Even if you don't buy chips, AI infra costs will likely rise as supply chain players will pass on costs   2. Entrepreneurs After decades of silence, there is massive opportunity in innovating memory Its still overlooked by many but we will soon see more high valuation memory startups which will become attractive acquisition targets for the 3 big incumbents   📷: FT   …   Did you find this helpful? ♻️ Repost this to inform your network 🔔 Follow me for more AI insights 🔖  Subscribe to my newsletter

15

4 Comments

At 3 AM on a Wednesday, Amazon needed me. I wasn’t even on that team anymore. Back when I was an SDE at Amazon, a Sev-2 had hit one of our services. Many customers were impacted. The team had been chasing it for hours. Still, they called me. I rolled out of bed, opened my laptop, and joined the bridge. The tricky part was that the fault started a week earlier. A code change had shipped behind a flag, traffic ramped slowly, and the failure signature was noisy. Logs were chatty in all the wrong places and silent where we needed them most.Dashboards showed errors, but no clear line to the root. We did unglamorous things. We wrote a minute-by-minute timeline. We diffed deployments, configs, and flags across regions. We checked queue backlogs and retry storms. We sampled logs with new filters. We bisected the traffic ramp. We rolled back the smallest safe thing first. Then we added one log, in one hot path, and the picture snapped into focus. By sunrise, we had the root cause and a fix. The impact was contained. Customers recovered. What stayed with me was not the fix. It was why they called. They did not call because I am smarter. They called because, over months, I had done the boring work. I wrote runbooks when no one asked. I paired with juniors on their first on-call. I shared context so others could make decisions without me. I showed up when it was not my ticket. Trust is not built in one night. It is built in all the small days that come before it. Ownership is simple. If it affects your customer, it is your problem. Org charts are for payroll, not for 3 AM. If you are early in your career, remember this: → Document the weird corners when you discover them. → Add one useful log where it hurts, not twenty where it is easy. → Keep a living checklist for rollback, not a static wiki page. → Teach one person the thing you just learned. → Read your dashboards when things are calm, not only when they are red. At 3 AM, you will not rise to the occasion. You will fall to the level of your habits. Build the right ones while the house is quiet. That’s how you earn trust that lasts beyond teams and ownership that outlives org charts. A few days later, this moment turned into an accolade from the team..  a reminder that trust is built long before it’s tested :)

334

45 Comments

"If I get laid off tomorrow, I won't be able to find a job" is a fear many devs are feeling in this market, even though they have done solid work. I took my exit from Amazon after 5 years, even though it was the company I started my career with, and I went through the same strange transition period as I got out of Amazon’s ecosystem. Here is what I learned. At Big Tech, you attach your identity to a logo, which you shouldn’t do. You wake up one day and think, “I do not know software engineering, I only know Amazon engineering.” The fear feels real, but it is a story your brain is telling you, not the truth. A few realities that are easy to forget inside the bubble: 1. Tools change, problems repeat. Every company has its Brazil or Peru. Outside, the names are different, the core work is the same. People, trade-offs, prioritization. 2. Your skills are more transferable than you think. Distributed systems, debugging, on-call, writing design docs, working through half-broken requirements. These are valuable everywhere, not only in FAANG. 3. The market cares about more than LeetCode. Yes, you need to pass coding rounds. But people are getting rejected for zero soft skills as well. Clear communication, ownership, and calm under pressure are still hard currency. So what can you do if this fear is sitting in your chest today? Here is a simple plan. 1. Detach identity from employer Build something outside work. Sport, hobby, volunteering on the “corporate” side for a non-profit. You will see how much you know that has nothing to do with internal tools. 2. Rebuild confidence with small public projects Take the type of system you work on today and recreate a tiny version with public tools. Use AWS, Docker, GitHub, whatever fits. You are practising the same thinking with a different toolbox, and you can talk about this in interviews. 3. Create a boring preparation routine Do not stare at LeetCode and panic. Pick a small target. For example: one problem a day for 30 minutes, three days a week. Track progress. Accept that it feels horrible for the first two weeks. Everyone else is rusty at the start as well. 4. Talk to someone who has moved Mentor or ex-colleague. People who left after 4 to 10 years will give you a lot of perspective; it can be a very good thing. 5. Test the market before you are forced to Update your resume. Apply to a handful of roles. Take a couple of interviews purely as data points. Once you see interview loops again, the fear shrinks. You realise the bar outside is different from the story in your head. If you were smart enough to get hired, ship features, survive oncall and navigate that environment for years, you are absolutely capable of learning a new stack and doing good work somewhere else. Do not wait for a layoff to prove that to yourself. Start collecting wins outside the bubble now, so when change comes, you are moving from fear to a plan instead of from fear to free fall.

589

36 Comments

Amazon puts over 50 LPA+ on the table for the SDE II role in India. Imagine you are interviewing for SDE-II role @ Amazon and in your system design round, the interviewer asks: "When I pause a movie on my phone and later open it on my TV, Netflix/Prime Video starts from the exact second I left. How would you design a system so a user can always resume from where they stopped, across all their devices?" Btw, if you’re preparing for system design/coding interviews, check out our mock interview tool. You can use it for free here: https://lnkd.in/gpCn7t2T [1] What a naive design would do Idea: Store progress only on the device. Flow - The player keeps the current timestamp in local storage. - When you reopen the app on the same device, it reads that value and seeks to that point. Why this breaks down? - You switch from mobile to TV and the TV has no idea where you stopped. - If you uninstall the app or clear data, progress is gone. - There is no single source of truth when you have many devices. Just local storage is not enough for a company like Amazon or Netflix. [2] What real systems do: server-side progress service Idea: Keep playback state on the server, and let every device sync to it. Core pieces - A "watch progress" service with an API like  POST /progress and GET /progress. - A table or key value store keyed by user_id + profile_id + title_id (+ episode_id). - Fields like position_seconds, duration_seconds, device_id, updated_at, status (in_progress, finished). What happens while you watch - The video player sends progress updates every N seconds or on key events  (play, pause, seek, stop, app background). - These are small API calls that update the row for that title. - If position crosses a threshold like 90 percent, the backend can mark the episode as "watched" and maybe reset resume point to 0. What happens when you open the app on another device? - The app calls GET /progress for the current title when you land on the details page or hit play. - The server returns the last known position. - The player seeks to that timestamp and starts from there, often with a "Resume from 37:12" prompt. Extra details you can add - Store a small history, not only one row, in case of bugs or rolling back state. - Use write behind or batching so heartbeats do not overload the backend. - Use TTL or cleanup jobs to remove very old partial watches. [3] Handling tricky cases Multiple devices at once - Two devices might send updates at the same time. - The service should keep the latest updated_at, or prefer the active device id. Offline viewing - For downloads, the device tracks progress locally while offline. - When it reconnects, it pushes a final progress event that updates the server. Privacy and size - Progress data is small but grows with millions of users. - A key value store or a simple sharded relational table is usually enough, since the access pattern is "get by key, update by key".

219

15 Comments

Stop treating Load Balancers and Reverse Proxies as the same thing. In System Design interviews (and production incidents), the distinction matters more than you think. As we move from SDE1 to SDE2, we stop asking "How do I make this code work?" and start asking "How does this system scale and survive failure?" Here is the deep dive on two critical components that often get conflated. 1. The Load Balancer (The Scaler) Primary Goal: Availability & Horizontal Scaling. The Job: Spreading traffic across multiple compute resources to eliminate Single Points of Failure (SPOF). The SDE2 Nuance: You need to know the difference between Layer 4 (Transport) and Layer 7 (Application) balancing. L4 is fast; it forwards packets based on IP/Port without looking inside. L7 is smart; it terminates the connection, reads the HTTP headers/path, and routes /api differently than /images. 2. The Reverse Proxy (The Shield) Primary Goal: Security, Unification & Offloading. The Job: Sitting in front of backend servers to hide their topology and IP addresses. The SDE2 Nuance: SSL Termination. Decrypting HTTPS handshakes is CPU-intensive. A Reverse Proxy handles this heavy lifting at the edge, allowing your backend application servers to focus purely on business logic (and communicate via HTTP inside the private VPC). The Reality Check 💡 In modern architecture (and cloud environments like AWS ALB or tools like NGINX), the lines blur. We often use a single component to perform both roles simultaneously—terminating SSL (Reverse Proxy) and then distributing the request to a pool of instances (Load Balancer). But knowing which function you are tuning—and why—is what separates a mid-level engineer from a senior one. Questions for the network: Do you prefer handling SSL termination at the Load Balancer level or strictly on the application server for end-to-end encryption? #SystemDesign #SDE2 #SoftwareEngineering #Scalability #DistributedSystems #DevOps

66

1 Comment

SDE-2s ask, How do I build this? Staff Engineers ask, Should we build this at all? To reach the next level, you must stop being a task receiver and start being a problem refiner. Senior engineers focus on execution, Staff and Principal engineers focus on the business case. If you can save the company six months of engineering effort by proving a feature isn't necessary that is a bigger win than writing the most elegant code in the world. Start questioning the why before you optimize the how. How do you navigate work? Comment below 👇 #softwareengineer #career #growth

53

4 Comments

One thing your manager will never tell you Your promotion depends less on your performance and more on whether your team is actually growing. When I joined Amazon as an SDE-1, my team had 25 people. By the time I became Senior, it had 120+. Looking back, my “fast promotion” wasn’t magic. The org was expanding like crazy and growth creates opportunity. Here is the part people don’t like hearing: ❌ If your team hasn’t grown in years, your career won’t either. ❌ It doesn’t matter how good you are. There is nowhere to go. ❌ The only real path up becomes waiting for someone above you to leave. So if you want to grow faster: Choose teams that are growing. Not teams that are comfortable. Skills matter. Environment matters more.

1,039

21 Comments

When math draws a hard line for large language models A recent research paper by Vishal Sikka and his son Varin Sikka takes a very different approach to evaluating large language models. Instead of benchmarks and demos, it uses mathematics. The authors show that LLMs operate within a fixed computational limit. Once a task requires more computation than the model can perform during inference, two things happen: the model cannot reliably solve the task and cannot verify correctness either. Incorrect output becomes unavoidable. This applies directly to many “agentic AI” scenarios - long planning chains, multi-step decision making, global optimization, and autonomous workflows. The paper does not argue that LLMs are useless. Quite the opposite. Within their computational domain, they are incredibly powerful tools. What it shows is that some classes of problems sit permanently beyond what transformer-based models can handle, no matter how much data or training we add. AI can become a phenomenal accelerator - for code, analysis, automation, and knowledge work. At the same time, treating it as a universal reasoning engine or a fully autonomous problem solver creates real risk. The research is a good reminder that understanding the limits of a tool is just as important as admiring its strengths. You can read about in more detail here: https://lnkd.in/dqQBQZxd

12

When you join a new org or switch teams in the same, it is quite an unsettling and borderline anxious experience. Here are a few things that I did to ramp up faster. 1. I consciously made an effort to remain unblocked 2. I read a ton of code and its history, sometimes even unrelated 3. I asked questions about the past, the present, and the future 4. I took up relatively mundane tasks - cross-team work, tests, and docs 5. I extended a helping hand in whatever capacity I could 6. I proactively met skip-level leaders and managers to understand the vision Some companies do have a culture of pairing up with someone existing in the team, but even if you do not get a mentor, it is important that you still navigate the situation and ramp up as quickly as you can. The above list is not exhaustive, by any means, and hence you can always add things that you find helpful in your context. but the lowest common denominator is to show extreme intent and interest to get output and drive outcomes. Hope this helps.

996

25 Comments

6 months into my job at Amazon, I broke production and thought this was the day my career ended because the company lost a lot of revenue...This is the story of one of the biggest learnings of my career. I was on an ad tech team. I was trying to learn the ad pipeline, went a bit too confident with a change, and triggered a failure. For 7–8 hours, ads were not being served. The moment I realised what had happened, my heart just dropped. I pinged my senior, called it out clearly, and stayed glued to the incident channel. Here is what surprised me. No one shouted. The first questions were: - “What changed?” - “How do we restore?” - “How do we stop this from spreading?” We rolled back. Checked the data. Verified that the pipeline was healthy again. Only after things were stable did the real work begin. That was the retrospective. We walked through the timeline, and I explained exactly what I did. We realised there was a big documentation gap. Some critical behavior of the pipeline lived only in people’s heads, not in any design doc or runbook. The conversation was not “Why did you mess this up.” It was “How did the system allow this to happen and how do we make sure the next person does not fall into the same trap?” I wrote the COE (Correction of Error) doc. My senior manager liked how honest and detailed it was. He told me, “You made a serious mistake. You also did the right things after. You reported it quickly, stayed on the fix, and left the system stronger than you found it.” That has stayed with me ever since. A healthy engineering culture does not pretend failures will never happen. It assumes they will. So it rewards people who: - Surface problems early instead of hiding them - Take ownership of their changes - Care enough to improve docs and processes after an incident - Focus on learning, not on finding a scapegoat Looking back, that outage was expensive for the company, but priceless for me. It taught me that good teams judge you not only by the bugs you ship, but by how you behave when those bugs show up in production. You will break something important at some point in your career. The question is not “How do I avoid mistakes forever?” It is “When it happens, will I have the courage to own it and the maturity to learn from it?”

545

28 Comments

In India, people have a narrow mindset about their career. It applies to tech also. People care more about optics, less about building and experimenting. We love to judge others when it comes to career choices, stability etc. India is in a transition phase where we are slowly shifting away from service and manufacture based culture to a more modern product mindset culture - very slowly though. But we still have mindset from that old 1990s era where we have a very narrow vision of how a career should be which eventually transforms young workers into labours, not innovators. If you look at China, they are already competing with US when it comes to innovation. China made it possible by radically changing their education system and investing heavily in their startup and product infrastructure. They embraced big ambitions. The result is very clear - they have DeepSeek Alibaba Group Baidu, Inc. Tencent etc and they did not build it in one night, it took them decades. Without embracing uncertainty and experimentation mindset, you won't be able shoot for the moon. That will naturally come with instability which we don't value here. India is progressing in tech, but we are way behind when it comes to innovation - we might have some random exceptions here and there. Different companies hire here just because we are cheap labours. Everyone says that India is hub of innovation, a pool of talented folks who can create big tech etc. All these look good in theory. In reality, hardly any company (home grown or multi national) innovates here. Companies hire in India because they get people who get things done. So, it's more about execution than research here historically. Also, India is a low trust society in-general. Most of the companies born in India treat their employees with superiority and toxicity. If India really needs to become a global leader, tech is a good space to start with no doubt. However, the leadership mindset in overall Indian tech work culture has to radically change. A mindset that would encourage risk taking, embrace uncertainty, a mindset that would respect all point of views and won't see things from a single lens. A lot of us question whether our govt. is investing money or do we have the right education etc. Valid points. But do we ever ask questions on how to change the leadership traits in Indian mindset in general which would enable us to take bold steps?

13

Are software engineering jobs disappearing from the US job market? Short answer: No. They’re not disappearing, they’re reshaping 🌱 📈 Trend (2020→2025): ▪️ Software Developers: Steady growth | ~15% projected growth ▪️ Data Scientists/Analysts: Faster growth | ~31% projected growth ▪️ AI/ML Engineers: Fastest growth | ~38% projected growt What’s really changing: More roles now mix software + data + cloud + AI. Why it feels tighter: 2022–23 corrections + slower posting growth + higher skill bars. All three roles stay well above US median pay (BLS). Bottom line: it’s a skill shift, not a disappearance (yet 🤔) Sources: U.S. Bureau of Labor Statistics (OOH/OEWS) and major industry projections (2020–2025). #Jobs #Tech #JobHunt

23

2 Comments

🚨 60 day grace period for H1B workers who lost their job is no more Most software engineers moving to the US for a job do so with a H1B visa – you can get a L1 (but then your stay is tied to the company) or an O1 (but that's fairly restrictive). In the past, in case you lose your job, you could stay 60 days in the country to find another job. After that, you have to leave the country but can still come back if you get a new job and get your H1B transferred (e.g. you don't need to go through a lottery again) The issue now seems to be that the 60 days grace period is gone[1], so essentially you have to immediately leave the country once you're laid off. Being an immigrant always come with some uncertainty on your situation but this wasn't too much a problem in the past because: - Tech companies were not doing laid off - The job market was in favor of engineers – it was easy to just find another job - While 60 days is kind of short, it was still enough in many cases (or at least to sort part of your personal situation) This is a pretty big blow for the industry – while some (especially younger folks?) will still move to the US, less and less will be willing to do (especially if you have no short path to a green card) The situation is also even worst for people in the current situation that may be overstaying because their 60 days grace period was suddenly gone and may not be able to transfer their H1B visa after 🙁

73

20 Comments

How I reverse two-way door decisions as a Software Engineer at Amazon In a previous post, I talked about bias for action and two-way doors. These are decisions you can reverse if they don’t work out. Shout-out to Abhishek Chaurasiya who reached out via DM and asked me a great follow-up question: “What if you realize right after making a decision that it was the wrong one? What do you do then?” Here’s how I think about it: if you notice the mistake right away, you’re lucky. The longer you go down the wrong path, the more costly it will be to reverse. But the decision to reverse doesn't only affect you, so what do you do? 1. Stop and gather data. Don’t just say “I think this is wrong.” What evidence supports the change? For example, if you decided to make a service synchronous, but then realize the workloads are bursty, gather latency and throughput data to demonstrate why an asynchronous design is more resilient. Evidence will make reversal easier to defend. 2. Anticipate pushback. In a team setting, people may want to stick with the decision. Try to understand their motivations for wanting to keep the wrong decision - some are valid, some aren't (e.g. helps with someone's promotion because a system they own gets more usage). Be ready to explain why pivoting now prevents bigger pain later. 3. Reflect on the root cause. What led to the wrong call? Was it missing data? Rushing under pressure? Missing input from key domain experts? Reflection is how you avoid repeating the same mistake. If it’s your own decision, document the lessons. If stakeholders are involved, bring data and reasoning to the table. Two-way doors only help if you’re willing to step back and learn why you took the wrong turn in the first place. Trying to save face will make the problems worse. How do you handle it when you realize a decision you just made wasn’t the right one? --- Hello, I’m Charles, and I am committed to share my two cents on career-related topics for 100 consecutive days. In these uncertain times, I hope to support those facing layoffs or career challenges. Follow me, and let's navigate this together! (49/100) #BiasForAction #Leadership #DecisionMaking #CareerGrowth #SoftwareEngineering

25

In my video, I share some of my core learnings from my experience of switching, growing and getting promoted. What I cover: 1. How to grow in your current role? 2. How and when to choose your next role? 3. My preparation strategies as a SDE-2 @Amazon This video is for you if you are a college student, professional looking for a switch or growing in your current role. Link in comments 🖇️ #CareerGrowth #engineering #promotions #tech #fundamentals #Google #Amazon #ShareChat

64

4 Comments