Q1 2025 Update
Update on everything I learned + accomplished during Q1 of 2025!
The past three months have felt like a blur—one long sprint stitched together by calendar invites, late-night edits, and the occasional “holy shit, is this actually happening?” moment.
I wrapped up my junior year at UofT. Landed a summer internship at Shopify. Started working as a research scientist at the Vector Institute. Continued pushing toward my first paper submission to NeurIPS. Closed out my chapter at IRCC. Hosted an AI conference as one of the presidents of UofT AI. And somewhere in the middle of all that, launched a blog series called One Gradient Step at a Time—mostly as a way to make sense of everything I was learning before it slipped through the cracks.
It’s been a lot. The kind of good busy that still leaves you questioning whether you’re growing or just keeping up. Whether you’re chasing ideas or being pulled by them. Whether “momentum” is still yours to control.
I used to think progress was about stacking accomplishments. Now I’m starting to see it’s also about noticing what each one is quietly asking of you. Time, energy, attention. Curiosity, patience, presence. None of it is free. And lately, I’ve been thinking a lot about how to keep showing up fully without losing sight of why I started in the first place.
This is my Q1 check-in—what I’ve been building, what I’ve been learning, and what I’m still trying to figure out.
On Research & Understanding
Working at the Vector Institute has been one of the most intellectually humbling experiences I’ve had so far. It’s one thing to read machine learning papers, follow the latest breakthroughs on Twitter, or tinker with models in a coursework setting. It’s another thing entirely to wake up every day and try to push the boundary of what we know—even if just by a fraction of a percent.
When I first stepped into research, I had this quiet pressure in the back of my mind: the idea that real researchers come up with new ideas. Novel architectures. Bold hypotheses. Clever optimizations. The assumption was that the further you moved into research, the more original you had to become.
And while there’s truth to that—innovation does matter—what I’ve learned, and what I keep learning, is that novelty isn’t the hard part. At least not in the way I thought. What’s hard is understanding. Deeply. Patiently. Thoroughly.
I remember reading a quote from Ilya Sutskever a while back—something along the lines of: “Before trying to come up with new ideas, first make sure you understand the existing ones.” At the time, it sounded like solid advice. Respect the fundamentals. Know your history. But it didn’t really land until I started doing research myself.
Because here’s what no one really tells you: most of research is sitting with other people’s ideas. Not in a surface-level way, where you skim the abstract and jump to the results, but in the kind of way where you spend hours tracing why a particular loss function works, what assumptions are hiding inside a proof, how an architecture evolves across a paper series, or why one benchmark matters more than another. You realize that many of the “intuitions” floating around in ML aren’t actually intuitive at all—they're learned through slow, deliberate study.
In the beginning, I was eager to make progress. I wanted to contribute. I wanted to be useful. But I kept hitting this wall: I’d read a paper and feel like I got it—until I tried to build on it. That’s when the gaps would show up. Why did they do it this way? Why not that? How does this assumption hold up under different conditions? What happens when you scale this idea? Every question I asked created three more.
At first, it was frustrating. I felt behind. Like I wasn’t moving fast enough. But over time, I started to see that this was the work. Understanding isn’t a detour from research—it is research. It’s how you learn to ask better questions. It’s how you develop taste. It’s how you move from reproducing results to recognizing what actually matters.
I also started to notice something else: the researchers I admired most weren’t the ones constantly chasing the next big trend. They were the ones who could explain foundational concepts with clarity and depth. The ones who could walk you through the motivation behind a method, its limitations, the historical context that shaped it. They had range—but more importantly, they had roots.
That shift—from trying to be original to trying to be grounded—changed everything for me. It made me less obsessed with output and more focused on depth. It reminded me that every “new” idea is really just a remix of what came before, and that if you don’t understand the base layers, your additions won’t stand for long.
So much of machine learning research right now feels like it's moving at breakneck speed. There’s always another arXiv drop, another SOTA result, another flashy demo. But I’ve come to appreciate the quiet work behind the scenes—the reading, the replication, the debugging, the thinking. The part that no one sees but that makes all the visible parts possible.
That’s been my biggest lesson at Vector. Yes, research is about curiosity. Yes, it’s about asking hard questions and pursuing weird ideas. But more than anything, it’s about understanding the world you’ve inherited before you try to reshape it.
And that understanding? It takes time. It takes humility. It takes sitting with ideas long enough that they stop being someone else’s and start becoming your own.
Reflections From Finishing My Junior Year
Finishing my junior year at UofT didn’t come with the dramatic sense of completion I expected. No grand reflections. No sudden clarity. Just a quiet realization that something fundamental had shifted in how I worked—how I approached learning, structure, and effort.
In my first and second year, I planned everything. Hour by hour, sometimes minute by minute. My calendar looked like a color-coded command center: deep work blocks, study sprints, Pomodoro breaks, course schedules, meal prep—it was all there. That structure gave me a sense of control. It helped me survive the chaos. It worked… until it didn’t.
This year, I let go of a lot of that. Not intentionally, at first—it just happened. I found myself skipping the planning rituals I used to rely on. I’d wake up, look at what needed to get done, and just… do it. No time tracker. No neatly segmented blocks. And somehow, I was doing better than before. Performing better. Thinking more clearly. Feeling less burnt out.
At first, I thought I was slacking. That I was losing discipline. But the more I sat with it, the more I started to realize this wasn’t a breakdown in structure—it was a shift in internal calibration. I wasn’t working less—I just wasn’t clinging to control so tightly anymore. The structure had moved from my calendar into my instincts.
I could look at my day and feel what needed to get done. I didn’t need to map out every second because I had built enough self-awareness to know how long things take, when I’m most focused, when I need rest, and how to navigate the push and pull of motivation. I used to treat time like something to be conquered. Now I treat it more like something to be tuned to.
That shift also helped me see something else: not everything requires maximum effort. That’s not a license to slack off—it’s a recognition of scale. Some assignments deserve deep focus and late nights. Others don’t. One of the most useful skills I’ve picked up this year is learning how to right-size my effort. How much is enough? How much is too much? What can I skim? What deserves a deep dive? Where’s the line between rigor and overkill?
It’s a constant balancing act, and I don’t always get it right. But realizing that not every task is worth your full intellectual arsenal—that some things are meant to be good enough so that you can reserve your energy for what actually matters—that’s been game-changing.
We talk a lot about hard work. About discipline. But there’s another layer to it: discernment. The ability to know when to go all in and when to step back. When to follow the plan and when to trust your gut.
That’s what this year taught me. Not how to work harder—but how to work smarter. How to listen to my internal rhythms. How to let go of micromanaging my own time and start actually owning it.
Remembering the In-Between
One of the best decisions I made this quarter was starting my blog series, One Gradient Step at a Time. On the surface, it looks like a place to share updates—what I’ve been working on, what I’ve learned, where I’m heading. But underneath that, it’s really about something quieter: memory.
More specifically, it’s about not forgetting the mundane.
I’ve noticed something about how memory works—we tend to hold onto the extremes. The really high highs. The painful lows. Big wins, big failures, sharp moments of clarity. But everything in between? The regular days? The ones where you made slow progress, had small realizations, or just kept going despite the lack of momentum? Those tend to fade.
And that bothered me.
Because when I look back at a chapter of my life, I don’t want it to just be a highlight reel. I don’t want it to be defined only by the big things—the internship offers, the paper acceptances, the “I made it” moments. I want to remember the Tuesday afternoon walks I took to clear my head. The weird bugs I spent three hours fixing. The books I picked up and forgot to finish. The way it felt to be 21 and in it—not successful, not burned out, just somewhere in the middle, trying.
That’s what the blog is for. Not to impress anyone. Not to broadcast progress. But to make space for the parts of life that don’t get documented, even though they’re the ones shaping me the most.
There’s a certain kind of wisdom that only shows up in hindsight. But I think there’s also a kind that comes from capturing things before they’re fully formed. Writing when you’re still confused. Still figuring it out. Still mid-process. That’s what I want to hold onto.
And maybe one day—years from now—I’ll come back to these entries and realize how much of the story was written in the in-between moments. The ones I would’ve forgotten if I hadn’t stopped to write them down.
Closing Thoughts
Looking back on the last few months, it’s clear that growth rarely announces itself in real time. It happens in small shifts—in the way you think, in the questions you ask, in the standards you set for yourself without even realizing it. If you had asked me three months ago what I’d learn, I probably would’ve pointed to specific milestones: the research paper, the internship, the blog. But now, what stands out isn’t just what I did—it’s how I changed in the process.
I’m learning to slow down without losing momentum. To trade control for clarity. To see structure not as a cage, but as something that can evolve with me. I’m learning that research isn’t just about ideas—it’s about understanding. That effort isn’t about giving 100% to everything, but about knowing where your energy is best spent. And maybe most importantly, I’m learning to hold onto the ordinary days, because they’re the ones that quietly shape the story.
Thanks for reading—if you’ve made it this far, I appreciate you deeply. Writing these posts helps me stay honest with myself, and if it resonates with even one person out there, that’s more than enough.
Until next time,
Dev