Hongxu Liu

🚀 How to Approach a System Design Interview: Amazon Prime Video Example When tackling a system design problem, especially for a complex platform like Amazon Prime Video, it's crucial to approach it strategically. Here's how I break it down: Get clarity on the requirements, both functional and non-functional. Focus on the user flows, define the scale, detail out the components, improve scalability, and wrap up confidently. Pro Tip for Interviews: Don’t rush into drawing diagrams! Ask clarifying questions, understand the problem fully, and then build the system piece by piece. Communicate your thought process clearly, and focus on scalability, reliability, and cost-efficiency. Remember, the trick is to think aloud—explain why you're making certain decisions. The interviewer wants to understand your reasoning as much as the final solution! Good luck to all those preparing for system design interviews! You've got this!💪 Want to master system design interviews and land those $500k+ TPM roles? Make sure you're fully prepared with the Kracd.com interview prep course. Sign up today and transform your career in just 30 days. #SystemDesign #AmazonPrimeVideo #TechInterview #InterviewTips #Scalability #CDN #TechCareers #SoftwareEngineering #Microservices #TechLeadership

9

C++ Next Gen - I recently came across Herb Sutter's latest update on Cpp2 and cppfront. Check it out here: https://lnkd.in/gTabj6sx. I'm excited to see that the C++ next-gen - Cpp2 - is getting evolved. The Cpp2 is a new syntax of C++, which is compiled by cppfront. The compiler builds Cpp2 into standard C++ code. Impressively, C++ and Cpp2 are interoperability i.e. they can be used together like Kotlin and Java, Swift and Object-C. The cppfront promises to be safer and simpler. It introduces interesting features that C++ lacks, such as contracts, metafunctions, and reflection. Here's a snippet in Cpp2: main: () = {    words: std::vector = ( "Alice", "Bob" );    hello( words[0] );    hello( words[1] ); } hello: (msg: std::string_view) =    std::cout << "Hello, (msg)$!\n"; Although it differs from C++ syntax, mastering a new syntax shouldn't take much effort. Looking forward to writing in this new iteration of C++ soon! #cpp #cpp2 #cppfront

6

I 100% agree with this. In India, we are not focussing on precision level skill development. That's the reason all manufacturers are still dependent upon China due to dependency on parts which require precision manufacturing with efficiency. We easily say that China products are low quality, but if you compare the same cost products, you will see China's products quality is far ahead compared to Indian counterparts.

7

SWE job growth declined for the first time since 2000 The SWE market has had tremendous growth. Though now is the first time in two decades we’re seeing shrinkage in the number of job openings. Gergely Orosz wrote a wonderful, in-depth article on the state of the SWE job market. Learn more in the latest issue: https://buff.ly/4f5ldDe If you’re new to my content, every week I share a spreadsheet of recently posted, hand-picked software engineering roles to make your job search a little easier. To get the weekly jobs via email, subscribe here: https://buff.ly/4hkLOh0 Photo: Live Data Technologies

7

1 Comment

Today I'm sharing this 🔥 video from our recent Engineering onsite at Demandbase HQ in SF! https://lnkd.in/eFAFR6QG We were so fortunate to host Zhengxuan Wu and Aryaman Arora, two upcoming AI researchers from Stanford University Department of Computer Science. Their approach to fine tuning, called ReFT, targets the representations emitted by neural net layers rather than the weights of the layers themselves. This makes it very flexible when using the same base model for swapping multiple fine tunes. The memory overhead is also very low for their intervention weights. Another highlight is how powerful it can be. One POC they showed forced the LLM to speak in only emojis exhibiting ReFT's power to control output. Control is crucial when, for example, you want to prevent your bot from offering customers unauthorized refunds!! If you're already familiar with LoRA, it should be pretty simple to use with only a few extra hyperparameters -- details in the video. Here's a link to the repo. Give them a star! https://lnkd.in/eJScNbzK Thanks again for letting us host your first external talk Aryaman and Zen. It's very exciting tech!!

17

4 Comments

SWE -> MLE -> SWE During my career in tech, I transitioned from Software Engineering (SWE) to Machine Learning Engineering (MLE), and back to Software Engineering. One friend asked me what the key differences are in both roles. I thought a bit and here they are: 1. Longer feedback cycles in MLE. As SWE, I used to be able to test my work in minutes and see the difference. As MLE, the feedback cycles are way longer. It could be days and sometimes weeks depending on what cycle of work you are on (data cleaning, model training or evaluation). 2. MLE requires more thoughtful and intentional hypothesis driven development.  - Software engineers like to just try out stuff. When we encounter new problems or bugs, we like to come up with ideas and try immediately to see if it resolves the problem.  - MLE is a totally different beast. Because of the long feedback cycle, anything that MLE tries naturally carries more weight and is more costly (eg. time and compute). So you have to come up with all the hypotheses possible and pick the most probable to try. This is why good training (eg. PhD degree) and experience becomes much more valuable and visible in MLE performance. Because if you don’t come up with a good hypothesis, it could take you many iteration cycles to resolve a simple problem. 3. Reasoning and attribution.  - It’s way easier to reason about software systems through code. You can generally answer why things work or do not work.  - Whereas, traditional ML models all lack explainability. Even after hypothesis based iterations, performance attribution can be uncertain. Instead of trying new model architecture, most effective improvements often come from gathering more data for underperforming areas. As a result, I found it less enjoyable as an MLE and started transitioning back to SWE role. What has been your experience with these roles? Do you see similar differences? #SoftwareEngineering #MachineLearning #TechCareers #MLvsSWE

104

7 Comments

Here's my advice to a new junior engineer in their first two weeks on the job: Not surprisingly, the industry is nothing like academia, and that's OK. What advice would you add? What made your onboarding an easier experience? https://lnkd.in/gQ2-WZby

8

So, I'm actively looking for a new job (two weeks left before my layoff), and I just found an interesting method for generating cover letters. I've been using ChatGPT for a little while, by simply prompting it with: "Please generate a cover letter for me for the following job posting:" and then pasting the post. That worked pretty well, but I still had to do a decent amount of rewriting. I've tweaked this method a little bit, and am now getting much better results. I started a new chat, and began it with: > For the remaining conversation, I'm going to ask for cover letters for various job postings. Here, I'll give you my resume, and would like you to base the cover letters off the experience listed in this resume. Here is my resume: I pasted a plain text version of my resume after that. It responded with: > Based on your extensive experience, here's a tailored cover letter And then gave me a generic cover letter. But, after that, in the same conversation, I do something like: > Please create a cover letter for me for the following job posting, based on my experience: And I paste in the job posting after that. Now, the cover letters I get include mentions of my previous work. It's not perfect, by any means. I still have to check it for inaccuracies, and weird word usage (it really likes the word "enthusiasms"), but now I can get job-specific cover letters, with my experience included, with very little effort. #JobSearch #OpenToWork #AI

65

6 Comments

Gergely Orosz's recent article about why people leave Big Tech (https://lnkd.in/guPM_Z-C) is broadly on point. The article is very long, and methodologically lean (citing data from just "several" engineers), but IMHO the key themes hold: decreasing stability, more growth opportunities in other environments, closed or slower career paths, being forced out (way more common than LinkedIn would let you believe), etc. Ethan Evans had a similar point, saying that "Today there is a good argument that Amazon is more of a down escalator." I'll throw one personal observation into the mix: IMHO, working at Big Tech can be "high experience variance". Meaning: your opportunity to create value for the product or company, and also your experience while doing that work, can vary wildly. This is true even within the same team, let alone between teams or orgs. And of course it varies over time, both organically (e.g., as projects happen) and inorganically (e.g., reorgs, priority shifts). Of course, I am just drawing from my personal experience, so YMMV. My experience was not unidimensional (sometimes I grew technically, sometimes managerially, etc), and time-varying. For some people, high experience variance may seem obvious, but for others, it may be a surprise. Hence, I am sharing this thought for the latter group. If you think "I am joining company X, therefore I expect to do meaningful work, and ideally interesting work", I would suggest not assuming that. Maybe you will, and maybe you won't, sorry to say, it really, really depends. How to optimize for high value experiences is a topic for another time.

12

2 Comments

If you haven't seen the new oxc-transformer release, you should check it out: https://lnkd.in/e_2KYHWq This morning I released a new version of hohoro (a minimal JS/TS build tool for libraries supporting incremental builds) with a new experimental export and binary that uses oxc under the hood! This means you no longer need to install or configure swc and also tsc for library builds! Check out hohoro here: https://hohoro.vercel.app/ and give it a try in your next project! The coolest part of this - is that all you need for: - Building your library (.ts/.tsx/.jsx/etc ➡️ ESNext JS) - Generating types (.d.ts files) - Incremental builds (only building changed files) Is to: - Install `hohoro@0.3.0` - Add the script: "build": "hohoro-experimental" And that's it!

5

1 Comment

AI & SaaS: A Tale of Two Futures What if Excel and other SaaS apps don’t vanish but evolve, blending seamlessly with AI? In a recent Interview (https://reurl.cc/eGz06j), Satya Nadella, Microsoft’s CEO, suggests the “agent era” will replace traditional applications. That’s Nadella’s vision: AI as a co-star, enhancing our favorite tools—think Excel with Python instead of Excel replaced by something else. But here’s the twist: 🔸 I see AI’s disruptive power surpassing any need for standalone software. 🔸 Why open an app when a smart agent can directly pull, manipulate, and display data on demand? Nadella believes AI + SaaS = Harmony. He sees Copilot orchestrating tasks, quietly boosting productivity. Developers still matter, refining user experiences to work with AI rather than against it. I foresee a more dramatic shift. If AI can generate everything from sales forecasts to text summaries instantly, maybe the spreadsheets and CRMs we rely on will fade away. Where do you stand? 📌 Collaboration or Disruption? 📌 AI enhancing SaaS or AI replacing it? No matter which path emerges, one thing’s certain: AI is rewriting the rules—and the next chapter of SaaS will be far from business as usual. Your thoughts? Let’s discuss. #AI #SaaS #FutureOfWork #Microsoft 

4

These books focus on more than just coding—they teach how to think like a systems architect and solve complex challenges.📚💡

1

I found a highly useful book on how to handle "crucial" conversations, for example: how to provide feedback, how to disagree, share alternative point of vew properly: https://lnkd.in/gwjwfB4r It might be useful for someone else too.

13

After having conducted ~150 engineering interviews, here is my advice for candidates: 📚 Get your fundamentals right. I won't ask you about esoteric optimisation problems in a red-black tree. I will not hold it against you if you cannot explain the optimal algorithm to generate a Voronoi diagram. I do expect you to know Binary Search, though, and at least the most common data structures (i.e. List, Set and Dictionary) - and yes, you should be *somewhat* familiar with Big O. 👩‍💻 For coding: Ask clarifying questions that help you understand the requirements and establish the edge cases. If you make assumptions, make them explicit. Go for a simple, maybe even naive solution first - then iterate and improve it. Explain your train of thought. I am more curious about how you approach a challenging task rather than in how well you can reproduce the ideal solution off LeetCode in an interview situation. 💡 If you are looking for ways to impress me - come up with a smart tradeoff and explain your reasoning to me. Don't try to demonstrate your ability to handle overboarding complexity (unless you are really confident you WILL succeed). You getting stuck halfway into a super-sophisticated solution attempt suggests you tend toward building complex solutions for simple problems. You being pragmatic and able to realistically assess what you can achieve within the timeframe of the interview shows me you will deliver results once hired. 🤷‍♂️ If you don't know - just admit it. None of us have all the answers. I trust you know how to close those knowledge gaps quickly when the need arises. 🧘🏽‍♂️ Relax. I want you to succeed. I really WANT to fill that position, that's why I invited you for the interview in the first place - which is why I am rooting for you! I won't set up any traps for you. If you get stuck, it's ok to ask for a hint. Actually, I'd expect you to do exactly that once you got the job. #engineeringinterview #interviewadvice #techinterview

30

1 Comment

This morning I've got a guest post on Google's Open Source Programs Office blog: what to do when your open source contribution stalls out. There are posts and posts about how to improve contributor experiences from the maintainer side, but I couldn't find much about how to push through a rough contribution process from the contributor side. Turns out you, as a contributor, actually have quite a lot of power here! I wrote down some of the strategies that have worked for me and my team as we contribute to the Git project; they apply for open source work, but also for any kind of decision stalling out. https://lnkd.in/g3eD6W4N

26

2 Comments

🚀 Exciting Opportunity Alert 🚀 Are you a Principal Engineer passionate about Distributed Machine Learning? Check out this role in Santa Clara, CA: https://bit.ly/3Zl5M4O. Key skills: #gpu #apache #spark #machinelearing #ml #dl #deeplearning #acceleratedcomputing #cluster #rapids #mLlib #XGBoost #PyTorch #cuML #pytorch #tensorflow #apis #api #Kubernetes #YARN #Ray #C++ #Scala, #Python #nvtabiular #cuda #horovod #petastorm #dll. Interested? Reach out to RobKemp@Nvidia.com for more details. Opportunity awaits! 🌟 #engineering #machinelearning #distributedcomputing

3

🎄🎁 Advent of Open Source – Day 4/24: WenFire: When Could I retire?! 🔥 (See my intro post https://lnkd.in/g3Wfuid3) While I really love my job at IonQ, I believe I'll love it even more when my family's livelihood doesn't depend on it. That's the dream of financial independence - doing what you love because you want to, not because you have to. This led me to create WenFire, a financial independence calculator that helps answer the eternal questions: "When can I retire?" and "How many days of extra work will that new iPhone or car cost me?" 📖 Origin Story There are many retirement calculators out there, but they all made assumptions I didn't quite agree with (and I'm sure others won't agree with my assumptions either!). I wanted something that matched my specific views on salary growth and investment returns. Plus, I had been itching to learn more about building web applications, so this project let me explore FastAPI, htmx, and Vega-Lite while solving a personal need. 🔧 Technical Highlights • Built with FastAPI and Bootstrap • Interactive visualizations using Vega-Lite • Responsive and mobile-friendly design • Customizable assumptions for:  • Investment growth  • Inflation rates  • Salary increases  • Safe withdrawal rates • Allows parameter changes at certain dates • Calculate retirement impact of one-time expenses • All parameters are encoded in the URL for easy sharing 📊 Impact While this project only has 3 GitHub stars, its real value is personal - it helps me track my own journey to financial independence. It's also been a great learning experience in web development, an area where I had little prior experience. As a child I used to make and sell a couple of websites, but that was a very different time. 🎯 Challenges and Solutions • Learning "modern" web development • Creating intuitive visualizations for financial data • Balancing simplicity with customization options • Making reasonable default assumptions • Ensuring mobile responsiveness 💡 Lessons Learned 1. Financial modeling requires careful consideration of assumptions 2. Web development is a different beast from scientific computing 3. Sometimes the best way to learn new tech is to solve a personal problem 4. The best calculator is the one that matches your own assumptions Want to calculate your path to financial independence? Try it at https://wenfire.nijho.lt or check out the source code: https://lnkd.in/gXyHiBKv #OpenSource #Python #FinancialIndependence #WebDev #FIRE

11