Before joining Precog, I was sure that I would be an engineer. I didn’t have clear plans for the future, but I thought research wasn’t for me. This blog is about how Precog taught me how to conduct research and thrive in an environment of uncertainty.

How it started

It all started in at the end of 2-1 when Haran (a good friend of mine) asked me which lab I was planning to join. I told him, “There is no way I could do research. I’ll do a BTP or work on a software engineering project” and he was genuinely shocked. After some back and forth, he convinced me to at least apply and see if I liked it first before making a big decision like this. I applied, and after a thorough application process, I was in.

3-1 And then it began.

My first semester at Precog was dense and eventful. I was shocked at how research was so different from everything I had done before. So much of what I was doing was underspecified which meant I had to make decisions without immediate feedback on whether they were right or not. This uncertainty was very new to me, and while it was initially uncomfortable, I gradually learned how to navigate it.

My first project involved working with LLMs, something I had 0 knowledge about. I had to learn everything really fast and on the spot. We had surprisingly quick success with an initial draft being accepted at a workshop. After this, we had the task of refining and extending our work for the conference deadline the next month. After many late nights, balancing coursework and semester exams, we pushed through and made a strong submission to a conference.

3-2 Building confidence.

During spring, I started becoming more proactive with research. I started pitching ideas to members of the lab and getting feedback. I would often discuss 2 to 3 ideas per week with Shashwat Singh in the lab and we had this generator-verifier setup going on, where I generate ideas, and Singh destroys them… immediately (at least most of them). Apart from the open collaboration that goes on at the lab, another aspect of Precog which helped me grow were the weekly updates. I was initially very introverted, but the environment which weekly meets provided let me engage publicly with various ideas and topics which helped me break my introverted shell bit by bit.

During this time, I wanted to expand my knowledge of deep learning, and so I started to explore GNNs (another part of the lab that I love, is that you can do whatever you want, and PK will generally support whatever you’re interested in) with Akshit under the guidance of Prof. Charu Sharma. During this time, I started actively testing the ideas that I had and found that most failed. The project was going slowly, and had little progress for several months. At this point, I wasn’t even sure if I liked research or not, and was leaning towards ending the project since there was only a month left in the semester. Even during this time, PK was patient, and was okay with the slow progress. We finally caught a break, and found an idea which actually worked, and we pounced on it but this initial draft was rejected.

4-1 Entrenched in Research

After my (uninteresting) internship, we started to plan out the improvements we would make during the internship itself, and started executing as soon as the semester started. Tensions were rising since I hadn’t done original work which was recognized yet and I wanted to prove myself.

While I was working on prior work, I found interest in deep learning efficiency [paper], and interpretability [paper]. PK trusted me to manage these 3 projects all at once, and I wanted results on all. I was planning on submitting to two workshops, and while PK was initially against this, he let me do it anyway. The freedom PK gives his students is unparalleled, and it was clear that PK wanted me to succeed just as much as I wanted to succeed.

Results came in. All 3 drafts were accepted, the AAAI draft [paper] even had a reviewer give it 10/10. I was relieved that my efforts finally bared fruit but I didn’t celebrate alone; all of Precog celebrated with me. PK’s unwavering support coupled with exhaustive ideation with members of the lab produced this result and that kind of support isn’t easy to find.

How it ended

Even in 4-2, I ask for advice from PK and still exchange ideas with Singh, and I hope open discussion becomes a long term trend in the lab. Looking back, my view on research would’ve been extremely different without Precog, and it shaped how I view the world. I came in certain I wanted to be an engineer but after Precog I think I might just enjoy the uncertainty and ownership that comes with research.

After graduation, I will be pursuing a Master’s from Carnegie Mellon University (CMU) where I intend to continue research. A complete list of everything I’ve worked on can be found at my Scholar. This lab didn’t just shape my career, it taught me how to embrace vague ideas, new direction, and the messiness of discovery.