A messy (but honest) recap of what I worked on in 2025, what I learned, and what I want to do better in 2026.
OK, so after dreading it for 5 days into the New Year, I finally sat down to write this blog. I have been thinking of writing this from the beginning of December 2025, but writing and publishing does not come intrinsically to me ,evident from my incomplete project page, and a half-hearted commitment to the second half of my drone blog (spoiler: sadly the project is currently on hold, and I am unable to devote time to actively work on it).
I get easily distracted, so completing this blog would be a challenge in itself.
Anyways, hi , thanks for reading this convoluted mess. Before you go ahead, I want to set clear expectations: this is basically a recap of non-exhaustive things I did last year, both personally and professionally, and the lessons I learnt along the way. Please do not consider this general advice.
These lessons have been helpful to me; I hope they are to you too, but if not, you do you.
Ok, starting with the 2025 recap: I worked in a lot of different areas, including Finance, Drones, Climate, Computational Biology, training LLMs, and of course - all connected with AI.
When I started, my reason was to test how far I can stretch my knowledge bandwidth and still be useful. While I got to work on so many things, each of these required significant bandwidth, causing slow progress across all my projects and, in some cases, a complete shutdown. This leads to my first lesson.
Not all projects need to be done,especially long-term projects. Those should be strictly limited to 2–3.
This year, I got the chance to participate in GSoC and ESoC, which, to be perfectly honest, I still do not know how I managed to get into. People often come to me for advice on how I did these, and honestly, for me it was 90% luck and 10% work.
For GSoC, I liked the org and the work I was doing. I am still part of the repository and plan to continue contributing. I truly, truly believe we need to do as much as we can to mitigate current climate change. This area might not be as flashy as the latest GenAI trends, but often these machine learning models and energy-based models help save millions of dollars and fossil fuels by accurately predicting generation and in a lot of cases, save lives by accurately predicting weather.
For ESoC, I was working in a similar domain to the problem statement I applied for, and I got in (Detailed blog about this coming soon). This brings me to my next two lessons.
Be persistent in your efforts. Do not run behind short-term gains.
Otherwise, it will be difficult to be persistent in your goals.
I also published my first research paper at a NeurIPS workshop.NeurIPS 2025 Workshop on Generative AI in Finance And I can say with 100% certainty: this would not have been possible without my PhD mentor’s support at FSIL Lab. I personally owe a great debt of gratitude to him for helping me develop parts of my research thinking. I have been volunteering at the lab for more than a year now, only because of the incredible people I get to talk to ,which brings me to my next lesson.
I will be honest: when I joined the lab, it was out of sheer brand value I would get on my CV from being associated with Georgia Tech. But I was soon enamored with the intellect of some really smart people in the lab.
I think initially more than 50 people joined through a summer position, but only 3–4 are left. Again, this would not have been possible if I had not been persistent in my efforts, and also without the incredible people I work with across different domains.
The list would be incomplete if I do not talk about AI. This has been an incredible year for people in GenAI and LLMs. I remember Andrew Ng saying this was going to be the year of “agents,” and truth be told, it has been (for example: Claude Code, Codex, etc.).
I have refrained from using them. Please do not take me for an AI pessimist—I am one of the biggest proponents of AI being a productivity booster,but slowly I realised it is only helpful when you know what you are doing.
As I approach my final time in university, I have realised I am an incompetent coder without AI, and I have been trying to reduce my dependency to write code. I still do a lot of debugging and QA with GPT, but I always ensure I am the one writing code in the IDE.
Arguably, the best way to use AI is to know when it is giving out slop—and you can never know this if you have only ever seen slop. Which brings me to my next lesson.
This pains me to say, but what I should have learnt in my 4 years of a CS degree (in terms of coding) is going to take me far longer to fill, because I have gaps everywhere.
A lot of people might find this controversial, but I genuinely think there will be an epidemic of bad code (especially with these summer-of-code-style programs—people are now going to open low-effort PRs).
But that is only part of the problem. A lot of graduates (like me) are facing a new kind of issue: without AI, they go blank in IDEs and cannot write a single line of code (just search on Reddit and you can find hundreds of posts about this).
My only suggestion: spend some time writing code without it. That is the only way I know. It is incredibly hard to do. I have personally been guilty of still using it—but I swear, I do not know any other way, at least for me.
Now coming to the final lesson, and arguably the most important one that I learnt this year:
I have gained more than 5 kg. I stopped going to the gym, giving excuses to myself that I need to work hard—but in truth, this has caused my health to deteriorate.
And when I talk about health, I mean both physical and mental health. Trust me: no job, internship, or paper is worth messing up your mental health.
Fortunately, this has been a good year for me. There have been ups and downs, but I was able to navigate those moments because I was mentally and emotionally capable to do so. If you are unwell, please reach out to professionals or your friends and family. Good physical and mental health is much more important than anything else.
If you are reading at any point of time next year and think I am not fulfilling my goals, or you just want to know my progress, drop me a mail—I would love a reminder.
Have a great year, and thanks for reading. Hope you got something useful out of this.
Cheers!