So it has been a long time since I updated my website. Well, it has been a journey: six years. Recapping it all would take some time, but I have a few paragraphs to share what I learned from various stints and the people I am grateful for. I will leave what’s next for me for the next article. What is certain, though, is that an academic path feels like the only path that allows for the intellectual freedom I need at this stage. It allows me to aim high and plant seeds that might bear fruit in the future rather than pruning hedges (Price et al., 2025).
Learning through deployment at Wadhwani AI
Wadhwani AI was such a roller-coaster ride. I spent nearly two years there, primarily across two projects. The first project focused on automating the interpretation of a tuberculosis test called Line Probe Assay (LPA). I worked in a three-person team and had so much fun. The high ownership and clear scope made it easier to make the project a success, and that external success gave me much-needed confidence in my abilities at the start of my career. I couldn’t have asked for a better first project. I am also glad that it introduced me to Keshav and Dr. Rahul.
After that, I wanted to focus on agriculture, so I joined the Cotton Ace project. As part of a larger ML team, I worked on improving our deployed object-detection model by adding a rejection model. It allowed the system to abstain from making predictions on photos that were dark, blurry, or outside the intended setting, such as a photo of a car, and request another photo instead. This introduced me to selective classification and to how substantially the product logic changes once a model is allowed to abstain. Because the second project was more mature, progress was slower and required more approvals to avoid breaking existing deployments. It was generally less exciting, but it taught me something about the care required when changing a system that people already rely on.
On both projects, however, I loved seeing deployments in the field. Every visit surprised me because seeing how people use your work teaches you things you cannot learn from the lab. I am grateful to have met very kind people during my time at Wadhwani AI. I cherish the friendships I found with fellow associates Nikhil, Pulkit, Anmol, Chandan, and Makkunda. Jigar and Jerome mentored our team and were excellent at breaking down complexity.
During my final few months at Wadhwani, I also saw how deeply a company’s direction and culture are anchored in the people who hold executive roles. When Wadhwani went through an overhaul, with new people filling major executive positions, it felt as if I were in a new organization. I only realized the extent of that change much later.
Believing in the problem
Next came Rephrase AI. I was amazed by the concentration of brainpower at the organization. Whenever I was stuck on a training problem, my manager, Nisheeth, could look at the loss curve and predict that a particular function was preventing certain parameters from being learned. This was the first time I saw how an intuitive understanding of mathematics could help you see solutions that are otherwise far from obvious. It was nothing short of impressive how the team achieved seemingly impossible things by thinking from first principles and because they genuinely wanted to solve the problem of virtual avatars.
I joined Rephrase because I wanted to be closer to state-of-the-art research, live in Bangalore, and experience the excitement of a startup. But I also realized that being indifferent to the idea of creating virtual avatars made it harder for me to push. It taught me an important lesson: work on things you believe in and can imagine pursuing for years without becoming bored. I know the power of compounding, but short-term excitement should still be taken with a pinch of salt. Even after learning this, it can be hard to keep my inherent curiosity in the driver’s seat consistently.
Returning to study
After Rephrase, I came to Tübingen to continue my studies. I had always intended to get a master’s degree, but COVID-19 changed the course of my life in 2020. It was time to return to that dream. I was driven by both curiosity about the world outside India and, to some extent, a desire to return to research. Landing in Germany and immediately signing up for 150% of the suggested credits in my first semester wasn’t smart. Once again, I wasn’t giving curiosity space to guide me; instead, I was forcing myself to be “productive.” Maybe the need to feel productive results from a scarcity mindset? I am not sure, but it certainly makes you choose things you would not choose if you were comfortable being average.
The last two and a half years have been intense. I have grown so much as a person that it feels strange not to have moved earlier. Throwing yourself into a different country allows you to see the customs of your home country for what they are: customs. Moving teaches you so much about the world and yourself.
Living in Europe has also made me much more attentive to my health. In India, it was often an afterthought; I would order food through delivery apps to squeeze in a few extra hours of work. Here, being a student pushed me to save money and become more mindful of what I ate. I learned much more about my body, nutrition, and hormones, and I experimented with creating a startup that helped people better understand their bodies. My idea was to help people run informed experiments with their health, grounded in current medical research. It was a lofty goal riddled with numerous obstacles. Maybe I will revisit it. I am much fitter today than I was when I moved from India, and that is reflected in my routine blood reports from Apoorv Pathology whenever I visit home.
Learning what I need from a workplace
While in Tübingen, I also got to experience working at a traditionally large company. My time at Mercedes was a mixed bag. Life is certainly more organized at a large company than at a startup, but one problem I faced was the language barrier. My German today is at the B1–B2 level. During my time at Mercedes, when it was perhaps at the A2 level, I remember attending group meetings held in German. I could not understand the discussion, which effectively meant that I wasn’t in the meeting. Lunch conversations would also happen in German.
This experience was very disappointing, but there is no quick, logical solution: if team members are simply more comfortable in German, asking everyone to switch is not easy. I learned that I need to work in a team where English is the shared working language. I am still working on my German, but I can’t imagine switching to it for work where deep understanding and camaraderie are so important.
Finding my way back to research
To hone my skills in building for customers, I approached Dr. Wieland Brendel, and he offered me a thesis project on creating an AI-powered tool to help teachers develop lesson plans. The work was intended to become part of Germany’s new open-source, nationwide Adaptive Intelligent System learning platform for schools. We designed the thesis as a mixed-methods study with two phases: gathering requirements and evaluating a prototype built from those requirements. During the first phase, I conducted numerous interviews and used them to plan the tool’s primary features. Next came development, which involved many sessions with Claude Code and plenty of bug fixing.
Finally, I evaluated the prototype with and without AI support using a between-subjects study design. Each participant took part in only one condition, a design that typically requires more participants to achieve statistical power comparable to a within-subject design. One thing I learned was just how time-consuming it is to conduct empirical research. It also reminded me that sustaining the necessary persistence once motivation dips depends on an inherent drive to solve the particular problem.
My thesis clarified the kinds of questions I want to pursue. I learned that the domain matters to me as much as the method: I am not interested in building a startup simply for the sake of it. I am most drawn to applying AI to medicine and, more broadly, to scientific discovery. I am especially interested in domains grounded in measurable phenomena and physical reality—the kind of grounding I associate with physics. I want to work with just one goal. Understanding.
That is why I find myself following curiosity back to academia. I want to answer questions using the scientific method. At this stage, I wish for nothing more than to see some of those answers move the needle a few years—or perhaps months?—down the line.
Until next time. :)