When the Stars Aligned

A TLDR?

The best decision of my life – personally and academically. And I am thankful to every single person who made it so.

The Beginning

The first question that most people asked me when I joined PreCog as a Research Assistant was how I found out about PreCog. It was a funny story. I was looking through CSRankings in September 2023, trying to shortlist universities for MS in fall 2024. I was half-minded about it- join MS immediately or get more research experience before I apply. During this process, a random thought of checking the university rankings in India occurred to me. Then I proceeded to check (read: stalk) every single professor on that page (there were a *lot* of them). That’s when I stumbled upon PK’s research and PreCog. 2 years later, I still remember spending hours looking through his LinkedIn and the PreCog students.

Before I knew it, I had drafted an email and hit the “schedule send” for the next morning at 8 am. But later in the night, the imposter syndrome hit, and I decided to cancel it when I woke up in the morning. To my horror, I overslept and woke up way past 8 am. But to my 2nd surprise, PK had already replied!

Without thinking of the logistics of moving to a whole other city, I just agreed. Two interviews and two weeks later, on my birthday, I was offered to join the lab.

During my first conversations with PK, when I mentioned it being my first cold mail and getting the offer on my birthday, he told me how sometimes the stars align for things to work out unexpectedly. And I think that is a perfect summary of my time at PreCog 🙂

A rollercoaster low in research

Once I joined the lab full-time, I was on three different projects, all in completely different research areas – HCI (with Akshay), graphs (with Arvindh, Akshit, Balakumar in collaboration with Google DeepMind), and LLMs (with Rahul Garg, Anirudh). A few months later, I joined another project in social networks, which was a collaboration with ETH Zürich (With Prof Ulrik Brandes and Meher), which Kshitijaa initially led. I was on four different projects by the end of April. In June, I also took up a Teaching Assistant role for the Responsible and Safe AI NPTEL course conducted by PK, Prof Arun Rajkumar and Prof Balaraman Ravindran.

This was one of the first lessons I learnt. As an enthusiastic person who joined the lab with all the hopes and dreams of working on several different projects, at the end of the day, it’s always about the quality over the quantity. Beyond a certain point, you will quickly burn out from being involved in too many research projects as a first author. It shows up in different ways- missed deadlines, uncertainties surrounding all your work, and the feeling of not giving your best to all the projects.

By October, I had hit a roadblock. One work was desk rejected, the other rejected, and the third project stagnated without any results or directions. This was probably the lowest point of the rollercoaster.

A slow-moving rollercoaster

At this point, I felt like I had wasted several months of my time and resources that were given to me without any result. During this time, my research interest also evolved. It was probably for several reasons- the lab research projects were mainly in interpretability, model attacks, evaluations and explainability, and the discussions surrounding them always excited me. Being the teaching assistant for the NPTEL course helped me learn a lot and increased my interest in this domain exponentially.

For a long time, I had been interested in joining Shashwat Goel’s PhD project, the research idea he had mentioned when talking to all lab members before graduating and heading to PhD. I was pretty persistent about it, routinely following up with him on the status and the idea progression. Once I had his green light for collaboration, I spoke to PK about it. We had a long discussion on how all my projects had progressed so far and if it would really be in the best interests of everyone to start another new project. But in the end, he agreed, and I was really grateful for that.

My work with Shashwat Goel on model similarity, a collaboration with the University of Tübingen (Joschka Strüber, Ilze Amanda Auzina, Ameya Prabhu, Jonas Geiping, Matthias Bethge), kicked off and was the most rewarding learning experience.

I decided to let go of two of my previous projects. My team and I decided to rework the graphs project (thanks to PK for pushing us to work on it when all of us were disheartened). I was now taking up more work on the ETH project, and Anirudh had joined the project. Applications for Fall 25 were opening, and I was applying to close to 13 universities (it was a lot of work).

The 2025 rollercoaster high

Until the end of January, things were neither bad nor good. I wrapped up most of my university applications (which I believed were way too ambitious but went ahead anyway), and the work with Shashwat was in its final phase, headed towards an ICML submission. And soon, I started hitting the highs in the rollercoaster.

The first was when our work on Reddit controversial post detection was accepted at the BeyondFacts workshop at WWW (We later won the best paper award for it). We wrapped up our ICML submission with a paper, code, Python package and a HuggingFace tool. We submitted our work to the SSI-FM workshop at ICLR as well.

Later, I got accepted at UCSD, NYU Courant and a few more universities in Europe (there were a few rejects as well, of course, but I had learnt over the year how to handle rejections). Our work was accepted at the ICLR workshop and ICML (as a spotlight poster!).

All of this, in a span of the last 4 months, out of the 1 year 6 months!

The finale

In May, when my time at PreCog finally came to an end, I was finally happy with how things turned out. Research-wise, I was proud of myself for having persevered despite the initial failures and, in the end, had work that I was proud of.

I am eternally grateful to the lab super seniors Hithkul, Anmol (whom I never met in person but still gave me a lot of advice over Slack and GMeet), and especially Prashant, whom I could turn to at any point in time for anything (I think everyone in PreCog will echo this sentiment). To all the immediate seniors in the lab whom I interacted with, thank you for sharing your wisdom with all of us. To the lab juniors, who are some of the nicest and smartest people I know, you will make great lab seniors. And to all my friends I made along the way, in and outside PreCog in IIITH, I had the best time of my life thanks to you.

Finally, I am grateful to PK for giving me this opportunity, believing in me despite everything, and letting me find my way through research.

As I head towards Tübingen for my Master’s in Machine Learning, I know that a huge part of me will always miss the lab, but at the same time, I am excited for all the new experiences that will come my way 🙂