TOC
The Context
Sep 2026
Sep was an extremely, extremely busy month for me. I can slowly feel the heat around and it’s getting extremely difficult and very competitive here. I loaded myself with as many difficult courses as possible and started exploring unexplored territories, so everything takes a lot more time than usual to wrap my head around. I had to catch up on the prerequisites while doing my coursework, which is extremely difficult for me right now. The best part is that I’m exploring some unexplored domains. My main focus for the next four or five months, at least until December, is operating systems. the reason? just to have fun and learn new things. I’ve always wanted to go in depth, understand, and be comfortable with how the computers we use work internally. I think a lot of the design principles,db design etc we use today were inspired by operating system design, so I wanted to explore how an operating system is built, not just theoretically. Theoretically, I know it, and we learned it in our engineering courses, but I want to learn it practically by hacking the kernel and tinkering with it. It takes an insane number of hours every week to wrap my head around it, play around with it, and figure out what to change and where. That’s going to be the main theme behind the next four or five months.
I’m also going deeper into distributed systems, where I’m focusing on implementing these ideas and strengthening my fundamentals. Since I’m learning operating systems and networking at the same time, I’m trying to connect all these pieces and tring to strengthen my fundamentals. Fortunately, I found the research lab I’ll be working in for the foreseeable future. That lab mainly revolves around building systems that can scale agents. It’s basically a distributed systems lab, but the workload is more about agents. The lab also has a mix of very interesting people: folks deep into ML inference, core ML folks, core systems folks, and people like me somewhere in the middle. So I have to set aside at least two hours a day for my research lab, on top of coursework, which is also time consuming. I think it will be really fruitful experience in the end.
The best part about this month is exploring my favorite job, which is teaching. I fortunately got myself a teaching assistant position, and One of the interesting things I did this month was build my own teaching materials. They’re entertaining and give a lot of value to the reader, and I’m trying to dumb things down so that a large section of my class, who are undergrads, doesn’t get overwhelmed. I want to help them understand the concepts efficiently. This is something I’ve never done full-fledged, and I’m trying to make the most of this opportunity by experimenting with a teaching style that’s efficient and entertaining at the same time. I’m also improving myself operationally. That includes figuring out the class, getting the materials right, handling all the operational stuff, making sure things are graded, making sure people have the right access to ask questions, doing office hours, etc which I would usually outsource to others,now I am making sure I learn and do it myself. I’m also focusing on communicating efficiently in class.. I’m slowly getting more confident in the things I know, so that I can convey them efficiently to others. That’s why I’m taking every opportunity I can to talk and present my work, so I get constant feedback on how efficiently I’m teaching.
At the same time, LeetCode keeps bugging me, like I said last month. I couldn’t find time to practice it, and I underestimated how difficult the job market is. I have to spend at least an hour or two every day to just randomly apply to jobs, because you only get opportunities if you apply. That’s something I need to do next month. But I’ve loaded myself with an insane amount of workload and can’t take on anything more, so I’m trying to figure out time for it somehow. If I mess this up, next summer is going to be really difficult for me.
The biggest learning this month is learning to say no. For the last couple of weeks I’ve been overwhelmed, because I’ve said yes to things I could have said no to. I’m coming from India, where opportunities are scarce, so every opportunity I see here feels valuable, and I try to catch hold of all of them. But because of this, I’m starting to feel I can’t concentrate on the things I’ve already promised to do. I’m slowly realizing that doing few things very effectively with full focus is better than doing ten things half-baked. Even though the opportunities are very interesting to be part of, I’m now slowly starting to say no to them, because I’m loaded for the next four months until Christmas. I think I just need to focus and do what’s on my plate efficiently, since that will be more valuable at this point.
Also, I somehow finished my dev vlog. I struggled with it, but I pushed it through. yayyyyy
Interesting reads:
The more I dive into distributed systems, the more I understand the value of reading research papers. A lot of what we find online is just surface level, while research papers are very helpful for exploring the depths. There are also insanely cool blogs from official companies like Netflix, Uber, etc., but the problem with them is they come up with solutions without talking much about the history. They don’t say what problem caused the solution or what technology existed at that point, and most of the time they miss those points. That’s why when you talk to somebody, they just randomly start with sharding and all the terms they find in designing data-intensive systems. My goal now is to read historical papers from the 90s and 2000s onwards, which were very influential behind the technology.
This helps me connect the dots efficiently, so that’s the side quest I’m focusing on. There’s just too much yay noise around, and the whole reading experience feels a little bad right now. So I want to pivot to influential research papers, which are evergreen. I’m learning a lot from them. Unlike today’s papers, each one is very dense, around eighteen to twenty pages with a lot of content, and exploring one paper takes me about a week. I’m documenting what I read every week here, so you can take a look, and if you want to read the same paper, you can always pick it up. paper readings
Books I read this month:
As for books, unfortunately I wasn’t able to finish any of them completely. But I started reading some books I got from my senior, and I started The 48 Laws of Power. I’m reading one law every day, so hopefully I’ll finish it in forty-eight to fifty days. So far it’s not great. It feels like an evil book, but I’ll keep trying to finish one law a day. Since I have a few books in reserve from him, I’ll keep reading after that.
I also started rewatching anime, so every day I watch one episode, around 23 to 24 minutes. Some days my book reading time goes to that episode. I really enjoy it, and it’s currently the only relaxation I have. I deliberately wanted to add it to my life, because otherwise things are already bad, and I don’t want my mental health to go for a toss.
So the theme of this month is that I wanted to be more ambitious. I signed up for a lot of stuff, and I understood that’s not the way to go, because it feels super overwhelming. So, starting next month, I’m trying to cut things off and focus on the stuff I originally wanted to do, so that I can barely survive the next few months. The goal for next month is to build the habit of applying to a lot of jobs. I want to be in the interview loop and figure out what they expect in this market, especially in America, so that I can start preparing accordingly.
Oct 2026
October is one of the hardest months in terms of workload and the number of hours I need to put in. Typically, if you put in a lot of hours, there’s an expectation that you’ll succeed. Despite putting in crazy hours, I miserably failed at a lot of stuff because of unwanted circumstances that got involved. I’m trying to eat a whole mountain within a few months. Digesting operating systems in such a small time, and the Linux kernel at that, takes an insane amount of effort and time, and that’s the only thing you can do if you want to do it in such a short period. My ultimate goal is to build the intuition to build the Linux kernel by myself. So far I’m halfway through, and conceptually I’m in a good place. I’m able to reason out why things are built in a certain way, how they can be better, and what the key learnings from the source code are. I don’t fear kernels anymore, but at the same time I’m not 100% comfortable navigating the Linux kernel source code. If you just give me the source code, it’ll take a while for me to find the right place where things are happening. So even though conceptually I clearly understand how Linux implements it, what the faults are, and why it does it this way, code wise it’s taking a bit more time to navigate and be extremely comfortable. I think over the next couple of months I’ll build up that intuition and the confidence to navigate the Linux kernel comfortably.
One major reason I’m taking a lot of time is that my kernel development setup is not that great. I have an old PC, and I can’t run QEMU on my local Fedora machine and build it, because I have neither the memory nor the space for a local deployment. So I have to boot a GCP VM and do all the development there. Sometimes the kernel crashes halfway through, so every time I need to take a snapshot, load the bootloader with the latest build, and then open it up and work on it. A lot of this was really manual work, and over the last week I’ve started to automate it using kexec and a lot of makefiles to navigate through GCP.
Research wise, I started this month really strongly, but towards the end it slowed down. A lot of my work involves going deeper into GPUs, and I don’t have a lot of time to explore those parts, because only now am I getting comfortable with CPUs and CPU abstraction. Maybe at the start of 2027 my focus will shift towards GPUs. Since my research involves both CPUs and GPUs, it’s taking a bit more time to build a concrete understanding and come up with solutions to the problem statement. So it has slowed down a little bit. Research should be my primary focus, but because of the crazy workload it’s taking a back seat. Somehow I’ve got to balance both. Another aspect I couldn’t focus on much this month is implementing distributed systems fundamentals. Since operating systems took away that time as well, I’m not able to sit and implement algorithms like Paxos or two-phase commit. I need to somehow buy time, but I’ve got to manage time equally.
I did LeetCode for about two days, and then I realized I don’t have to do it in C++ anymore, because it doesn’t make sense. I’m not writing C++ full time. I should be, but I’m not, and I don’t foresee myself writing C++ for at least the next few months. I can foresee myself writing Rust sometime soon, but not C++. My C++ knowledge is only competitive programming knowledge, and these guys keep bringing in new syntax every single time. The language has moved so much that it looks extremely ugly to even reason out what’s going on. The syntax is getting weirder and weirder. Since I’m not doing competitive programming anymore, there are no time sensitive runs. In competitive programming, even if an algorithm works in both Python and C++ with the same time complexity, there are times when C++ passes and Python doesn’t. Also, for competitive programmers, the C++ standard template library is so good that you can do almost anything with it. But you don’t need that for interviewing. At the same time, I can do competitive programming data structures in C++, but I can’t sit and write classes, abstraction, encapsulation, inheritance, and all the OOP stuff to write clean code. It takes a lot of time and effort to write all those abstractions and proper production code, and in an interview you just have an hour, so you can’t spend it all on these things. You’ll never get into the meat of the problem. So I thought, okay, let’s just do Golang from now on. But I’m not writing Golang every day, since I don’t have a full-time job, so I keep forgetting the syntax. Eventually I fear that even though I understand how the language is built and what its features are, without repetition I might forget the syntax. So I thought, okay, let’s start doing stuff in Golang. But I stopped completely this month because I was absolutely busy getting through coursework and midterms.
For the last two or three days I’ve been spending my time building a cute little personal project. It’s been so long since I built anything for myself, so I just wanted to build something I always wanted. My idea of a project is always the same: I need a strong use case that I’ll use every single day, and only then do I feel the urge to build it. This is something I badly wanted, so I sat and built it over the past two or three days. For the first time, I also wanted to challenge myself to not pay anything for infrastructure, figure out what free parts are available, and stitch them together to deploy. I feel ashamed to say this, but it’s the first time I’ve moved away from bash for my infrastructure setup and teardown to an infrastructure as code tool. I thought of using Terraform, but after thinking a bit, I felt Pulumi was a better tool for me because I could continue writing in Golang, and it’s such a relief to use Pulumi and not write everything in bash. Hereafter all my infrastructure as code is going to be either Terraform or Pulumi. It’s easier for me to review the code because agents can write the Pulumi files and I can go through them and see if everything is correct. Since coding agents are really good at writing code, it’s easier for me to get into infrastructure as code because I’m not writing any of it, I just have to review it. That’s a little less painful.
I’ll write a detailed blog about my stupidity with the system design, just because I wanted something for free. That will be a separate blog, next.
I started my second teaching job at the end of this month, and I’m really excited about it because it’s about one of my favorite areas, building a startup, and building a startup with AI. It’s an intersection of two of my interests, so I’m really looking forward to it. It’s an addition to my already existing schedule and it’s on Saturdays, so now I have even less time than I used to. I need to plan better.
paper readings is going at a decent pace. I covered a couple of papers this month, and I read a few more that I didn’t cover, because I want to cover only the papers that have something interesting or something I learned from them. That’s why I’m very picky about which papers I cover as part of paper readings. I’ll continue paper reading as long as possible, because I’m having so much fun and learning a lot more than from reading blogs. The content is extremely dense, and I don’t think it’s available online as part of blogs.
Interesting reads:
You can go to the paper readings and see the papers I covered this month. Those are very interesting.
- collecting favours by sunil pai https://sunilpai.dev/posts/collecting-favours/ is such a good read. It’s definitely a must read. It talks about how hospitality compounds. Just go read it, and I hope you feel better, just like I did reading it.
- after ai there is no average by sidu ponnappa https://sidu.in/essays/after-ai-there-is-no-average.html is an interesting take on how the pay scale gap between different types of employees changes and how organizations might start structuring themselves around power users of AI.
Also, this is not a read recommendation but a YouTube channel recommendation. If you’re Tamil and you understand Tamil, and you want to follow what’s going on with AI, there’s an underrated channel called Solvanam. Just go follow these guys on YouTube and regularly watch their videos. Since I’m chronically online and it’s my job to keep myself updated, I might have access to a lot more information. But almost 50 to 60% of what I consume is of no use, and it takes a while to filter out the remaining 40% that’s useful. These guys do a really good job of curating the best possible AI updates from that particular period of time. So check them out.
My interest has shifted from reading random blogs to research papers, but now I have a proper setup, which I’ve built over the last 3-4 days, where I can consume a lot more blogs. So I’m very interested in reading new voices.
Books I read this month:
Book wise, last month I started reading 48 Laws of Power, but I had to drop it because the content doesn’t align with my values at all. It feels super evil, so I just had to drop that book. This month I started a book called Book of Elon, which is curated quotes and interviews from Elon Musk. The author curated just the words of Elon Musk from his quotes and interviews, bracketed them into various topics, and packaged them as a book. I’m also halfway through Operating Systems: Three Easy Pieces and halfway through Linux kernel development volume 3, and hopefully in the next two months I’ll finish both. The problem with Linux kernel development volume 3 is that it’s a very old book. 30 to 40% of the ideas in it still hold, the principles of Linux development, but the Linux kernel is evolving, so the code base is evolving too. I’m having difficulty following the book because I need to go through the changes in the kernel and compare them with the book to see which is correct and which is wrong, and then read accordingly. That’s taking a bit more time.
Overall, I barely survived October. There’s a lot of hard work to do. In November I have to start putting in long hours every single day. I think that’s the only way I can get through the next two months. I’m happy that I started working harder after a month, and I have to double that in November. Now that I know exactly which areas I’ll be focusing on, I need to keep focusing on just these areas and try to get better at them.