Experiences,  StudentGiri,  Students

Precog: a low risk, high return investment

Precog has been a constant throughout my bachelors at IIITD. Having spent three-plus years at Precog, it is much more than a work experience or research experience on my resume. It is what has defined me and my work interests. In this blog, I will take a shot at the impossible task of summarizing my experience, learnings, and the indelible effect PK and Precog have left on me. 

Before joining my college, I was very well aware that there existed this majestic lab headed by this mystical professor. My first introduction to Precog was through my brother when I was still in 12th grade. He had applied to Precog but got rejected in the penultimate round. He showed me how well this lab’s alumni were doing, and I was sold. One of the major reasons I had for joining IIITD was Precog. So I was very well aware of PK and his lab and made it my goal to join the lab, ready to give whatever it would take. I was in my second semester, honestly, with not much on my resume, and I saw that Precog was accepting students.

PK’s Mail for Precog application

Precog was notorious for its rigorous admission process. I had to go through four rounds of tasks and interviews to get in. The most challenging was the programming task, which was to develop a movie recommendation system from scratch, hosted online within five days. As a second-semester student, this was a huge challenge for me, which turned into my first of many to come all-nighters. I learnt flask, collaborative learning algorithms, and hosting within this short span of time. But it was all worth it, and I got in. I remember it felt so good, because I wanted to get in so bad!

Getting into Precog

I started working with Vani Agarwal, and helped her with her thesis. I developed a flask application for one of her trained models. It wasn’t a lot, but enough to make me feel good about completing my first task. On the day of her defense, everyone was in the lab. That was the first time I witnessed how Precog was more of a family than a research lab. It felt like a family function, we had food and I got introduced to a lot of existing members, who in my mind were all “Pro”, having worked on something really cool (I had stalked each member enough to know what they were working on). 

First output in Precog (Acknowledgement in Vani’s thesis)

Precog had an elections “war room”, where the elections team would sit together, collect Twitter data and do a daily analysis. This election wide collection in 2019 resulted in 45 million tweets, which called for a lot of interesting analysis. This became my first project. I had endless fun working with Saurabh, late nights and it never really felt like work. And this is how I got my first publication. It obviously felt great but more than that gave me the confidence that working on interesting problems would naturally result in a tangible output. At this point I had become a part of the family. Lab became a place I would enjoy going after my classes, be it to catch up with other members, give updates on the project or receive gyan from Hitkul and other PhDs. At this point a couple of my other friends also got into precog. We would meet each week for the whats ups, which were really helpful. To be honest, many times I would not have any update, but looking at others working and producing output used to motivate me to work harder.

Precog’s Birthday celebrations

Then, came the pandemic and obviously things got online. But we still had fun. We had regular meets to watch videos, have lunch together or to discuss best time management strategies. I started working on my BTP “Analysis of masked and unmasked images from Instagram”, which would go on for two and a half semesters. This is when I realized how important it was to take ownership of your project. PK used to say “ Treat your project like your baby”. Many times during the project, we were at a place, which seemed like deadends, but they were local minimas. Instagram did not have an API we needed, we built a scraper, the scraper caused a server ban, we used Google colab, Google Colab would time out after 8 mins, we used proxies. We spent 8 months back and forth with the paper review, but it finally got accepted. It is the most satisfying feeling to drive a project yourself from start to completion. PK was a constant support, who would always back our project decisions and make us push harder every meeting.

PK’s Message after paper acceptance
Koo office, Bangalore

After my BTP, PK asked me to join the Koo thread. The project had started from a general comment on our slack channel. Koo, which was an upcoming microblogging app, interested us to explore what it was all about. We started collecting data, and performed some initial analysis, which got attention from news media like Print, Business Standard and The Hindu. Chirag pointed out  two days before the deadline, that the Asonam conference was accepting submissions and the entire team spent two days and two nights putting together a paper that would be awarded the “Best Student Paper” award. While working on this project I was so lucky to meet some great people, with whom I wish to work again in the future, but this was not the end, for the thread. Our analysis interested Koo as well, and they invited us to work with them in Bangalore. Having no offline internship experience, thanks to the pandemic, this was the best and the most satisfying experience I had working, in quite a long time. PK had shifted to IIITH at this point, and I never really got to interact with him a lot before this.

I count myself really lucky to be able to spend multiple weeks working very closely with him. It was so impressive to see him so motivated and committed to this project, being a professor. We would spend the entire day working, making presentations, and would then discuss the to dos for the next day before bed outside our rooms in the hotel. During this time, I was simultaneously working on the Elections thread with Professor Joyojeet as well. I recall a night when I was so tired, that I dozed off on a meeting with Jivitesh where we were supposed to code together, and would continue to sleep in the meeting for 4 hours. But it was all worth it, I was lucky enough to have insightful conversations with PK and even got to hear his life story over multiple dinners, which really left a mark on me. We worked on multiple problems during our time, ranging from community analysis to Sub Topic recommendation, and got to interact with the ML, Operations and Data heads on a daily basis. During this internship, I got to meet two really amazing friends in Anmol and Arjun, and I wish to have spent more quality time with them. The internship was a success, but what really made it special was the mail I received from PK after reaching home.

PK’s Mail after Koo internship

I continued working on the UP elections thread and Mental Health thread with external collaborators. The UP elections analysis got featured in Business Standard, Hindu, and Aaj Tak Radio and the results were published as a short paper at COMPASS, with multiple blogs to its lead up.  Precog provides a leverage no other lab does, giving you an opportunity to showcase your work, through channels like news media, twitter and publications. The mental health project also concluded as an ICWSM Dataset paper, which was an important milestone.

Publications as a part of precog

In conclusion, these three years have been an unforgettable journey. These three years were an investment I made in the lab, and an investment that PK, Precog made in me. In return I got more than I ever thought of : five publications co authored with PK, opportunity to work with and talk to external collaborators (Joyojeet Pal, Tavpritesh Sethi, Emilio Ferrara, etc),  timeless gyan, technical and  writing skills, friends, free food, T shirts and countless memories. And these returns compounded with every passing moment, making me realize the importance of starting early. With each passing moment, I was a bit more confident and clear in my mind than before. In return, I hope I left a good impact on the lab and was able to be resourceful in its building and its journey ahead.