It has been a while since I wrote here.
The main reason I did not want to write a new post is that a lot of the projects I have been working on are still in a very awkward in-between state. Some of them are at a standstill, some of them are still moving but nowhere near complete, and almost none of them feel finished enough that I can point at them and say, okay, this is ready to be written about properly. I did not want to put out a half-baked update just for the sake of saying I had posted something, and I definitely did not want to write one of those resume-disguised blog posts that are boring to write and even more boring to read.
But after sitting with that for a while, I realized there is something worth writing about even if the projects themselves are still incomplete. Since my tiny TPU work, the work itself has started feeling more real. Not bigger in the vague motivational sense, but more real in the engineering sense, because it is getting closer to tapeout constraints, compiler infrastructure, and systems that people actually want to use.
I think what has been bothering me is that I still have a very stupid habit of only granting emotional value to things once they are done. If something is unfinished, delayed, or stuck in a limbo where I cannot neatly package it into a result, my brain immediately starts treating it like it does not count. That mindset is probably useful for shipping, but it is also terrible for reflection, because so much of technical life is just spending months inside ugly middle states where nothing looks impressive yet and everything still feels uncertain.
And honestly that is where I have been living for a while now. Not in the clean ending of a project, but in the messy stretch where the work is starting to ask harder things from me than it used to.
I am also entering my final year of engineering now, which still sounds slightly fake when I say it out loud. It is exciting in the same way that all real transitions are exciting, because you can feel life becoming more concrete and your choices becoming more consequential. But it is also sad in a way that I do not think had fully hit me until recently. College is actually ending in less than a year. That is such a stupidly obvious sentence, but I do not think I had emotionally processed it until now.
And a big part of why that feels so heavy is SRA. Everything about me, at least the version of me that actually started building things seriously, came out of SRA. I started doing projects after looking at MARIO and Synapse32 there. Synapse32 itself has basically been with me for around two years now and was one of the first real things I picked up inside the club. TinyTPU, all the experience that came after it, the confidence to apply myself harder, the path that eventually took me to IIT Madras and then further into the work I do now, all of that traces back to SRA in some way. Even most of the people I would actually call my friends are from there. Looking at all of us now, and seeing everyone grow into their own serious work, makes me weirdly emotional because we really have come a long way.
What makes it even funnier is that SRA was not just some neat extracurricular line item for me. It was genuinely my second home. I hated lectures, and I definitely did not want my mom to know how much I bunked them, so I would just go to college early, disappear into SRA, sit there and do my own work, maybe finish assignments sometime in the afternoon, and then keep working or just hang around with friends. So much of my actual college life happened there and not in the places that were officially supposed to define college. That is why leaving it feels strange. We ended on a high note, we did a lot of cool stuff for the club, and we had a lot of fun, but it still feels surreal that we are slowly leaving the place that held so much of our everyday life together.
TinyTapeout changed how I think about hardware
One of the coolest things I got to do after that last post was send a 4x4 BitNet 1.58 implementation for tapeout through TinyTapeout. Even typing that still feels slightly unreal, and I even have a GDS viewer page up for it now. There is a big difference between “I built a thing and it works in my setup” versus “this is now going toward silicon, so every assumption should be treated with suspicion.” Tapeout has a way of making hardware feel less romantic and more honest, because you stop thinking only in terms of whether an idea is clever and start thinking much harder about whether it deserves to exist under real constraints. Area matters, simplicity matters, interfaces matter, and even small mistakes start to feel much more expensive than they did when the whole project lived inside the safe little bubble of local experimentation.
CIRCT has been raising my bar
Another major thing in the last few months is that I became a core contributor to LLVM CIRCT through GSoC. To be honest, I do not care that much about the label of GSoC itself. The useful part is the work. CIRCT is the kind of codebase that forces you to stop hand-waving, because you cannot just have a neat idea and call it a day. You have to make the idea survive review, fit into existing abstractions, interact correctly with the rest of the compiler stack, and stay readable for people who were here before you and will be here after you. That has been great for me.
Before this, a lot of my learning came from building things bottom-up: writing RTL, testing on boards, and trying to make one project work end-to-end. CIRCT has been a different kind of education. It makes me think more carefully about representations, transformations, lowering paths, and what it means to build infrastructure instead of a one-off demo. It also changed how I think about hardware tooling in general, because good hardware work is not just about writing better RTL. A lot of the leverage sits one layer above that, in the tooling and compiler flow that shape what hardware people can express, optimize, and maintain.
I think the personal part here is that contributing to something like CIRCT also made me confront how much I still do not know, but in a healthier way than before. A year ago that would probably have sent me into one of my usual spirals of feeling behind everyone. Now it mostly makes me feel excited. Intimidated too, obviously, because there is no way to work in a codebase like that and not feel small sometimes. But it is the good kind of small, the kind that reminds you there is serious craft here and that you get better by staying in the room long enough to absorb it. That has been one of the most important mindset shifts for me this year.
Systems work feels very different when it has to become a product
I am also currently interning at zett.ai, where I am working as the system architect for an on-device LLM box called Kimibox. This has been another category of “oh, this is real now.” When you work on something that is supposed to become an actual product, your thinking changes again, because the model is not the whole story, the accelerator is not the whole story, and even the software stack is not the whole story. The whole box matters. You start caring much more about system boundaries, movement of data, deployment friction, memory pressure, transport, reliability, and all the boring-seeming glue code and architecture decisions that usually get skipped in flashy demos.
And honestly, I like that a lot too. I have always enjoyed projects that cut across layers, and this work has made that preference even more obvious to me. I do not just enjoy isolated hardware or isolated ML or isolated compiler work. I like the full-stack systems mess where each layer keeps putting pressure on the others, because that is where things start getting interesting and where engineering starts feeling less like a demo and more like a real system.
It has also been weirdly clarifying on a personal level. A lot of student engineering culture rewards the clean demo, the exciting benchmark screenshot, the project that can be explained in two lines and looks impressive immediately. Product work is much less flattering than that. You can spend a lot of time thinking about constraints, failure cases, interfaces, or deployment details that are not sexy at all, and yet those are exactly the things that decide whether a system is real or just cosplay. I think I needed that reminder.
Maybe that is also why this post took me so long to write. I kept waiting for some grand sense of completion that never came. But maybe this phase of my life is just not about neat endings right now. Maybe it is about learning how to stay patient inside unfinished work without treating that unfinishedness as failure.
What I think changed
When I zoom out, I do not think the headline is just that I “did more stuff.” The more important thing is that the work is starting to connect. The tapeout work taught me to respect hardware constraints more seriously, CIRCT is teaching me how hardware tooling and compiler infrastructure shape the whole design process, and Kimibox is pushing me to think like a systems person building around real deployment constraints instead of only chasing a technically cute demo. That combination feels meaningful to me.
I still have a lot to learn, a lot, but for the first time in a while I feel like the different threads of my work are not random anymore. They are starting to point in the same direction, and I think that direction is simple: I want to keep getting better at building real systems across the stack. Not just projects that look cool from the outside, but systems that survive contact with constraints, other people, bad assumptions, and the long boring middle where most of engineering actually happens.
Another thing that has changed this year is that I feel a little less tightly wound than I used to be. I have always been too serious about everything in life, and in a lot of ways I still am, but my friends in SRA definitely chilled me down. I think they helped me realize that not everything in life needs to be squeezed so hard for meaning or treated like some high-stakes test of character. Sometimes life just deals you good cards on its own, and your job is to do what you like with the people you like spending your time with. That sounds obvious, but I do not think it was obvious to me for a long time.
I also think I have become much more comfortable admitting that I am actually good at what I do. After JEE, I genuinely did not think I would be capable of doing a lot of the things I do now. Even on the software side, where I always felt slower than I wanted to be, a lot changed for me once AI coding tools became part of how I work. Not in the fake “AI replaced learning” way people dramatize online, but in the very practical sense that it removed a lot of friction between what I could think through and what I could actually build quickly. If that had not happened, I honestly might have drifted even more aggressively into pure hardware work and kept software as something I tolerated instead of something I can now use properly.
Outside of work, life has also had these small details that make the phase feel more real. I got a pair of Audio-Technica ATH-M20xBTs recently, which are basically entry-level studio monitors, and I have been enjoying how flat they sound. It has weirdly made me appreciate a lot of older digitally recorded music more, especially songs where the drums and beats are less punchy in the modern exaggerated way and you can hear more clearly how deliberate the production choices were. I have also been taking more late-night walks, reading more peacefully, and spending time with compiler books too. I am currently going through Douglas Thain because ever since I got interested in compilers, I have wanted to build that side of my understanding more seriously.
I think what I am most grateful survived all of this is very simple: friendships, ambition, and curiosity. My taste has definitely gotten better, I am learning new stuff every day, and as long as that continues I think I will stay pretty satisfied with my life. So I guess this post is less “look at all the things I did” and more “this is the kind of person and engineer I think I am slowly becoming.” That feels like a much better update than “here are four bullet points from my life.”
cheers.
