Pranjal Garg

Pranjal Garg

I study how biological and artificial systems
learn, adapt, and reorganize through interaction.

About

There once was a flat sheet of cells,
That were stumpy and ugly as hell;
But one day they arose,
Stood tall on their toes,
And declared they were the best cells of all.

— T. W. Sadler, Ode to a Placode

Hi! I am Pranjal, from a small town in North India. I am an undergraduate student in Data Science at IIT Madras. Before this, I studied medicine (MBBS) at a premier public institute in India, AIIMS Rishikesh. I dropped out of medicine for various personal and professional reasons. Some of them are listed here. I am interested in two broad intersecting questions:


How does intelligence emerge from substrates?

I am interested in the organic basis of intelligence, in how it emerges from the interactions of neurons, astrocytes, the tissue microenvironment and the world an animal lives in. I have studied the visual system of Drosophila, mapped sensory neurons in S. hermaphroditum, modeled how astrocytes shape network activity, and analyzed the dynamics of hippocampal and orbitofrontal circuits. Brief overview of my work in neuroscience.


How do artificial intelligence systems work and how can we make them safe and reliable?

I am interested in mechanistic interpretability from both perspectives, curiosity-driven and pragmatic. As AI systems become more capable, we need reliable ways to check what they are doing. I am learning interpretability, monitoring methods and alignment through a top-down approach. This means testing when a probe or monitor is truly reading the model and when it is misleading us. I also study how models behave when their inputs are corrupted or their evidence conflicts, since those are the situations where failures matter most. Brief overview of my projects in AI.

Currently

I am pursuing a BS in Data Science and Applications at IIT Madras. I will be visiting VIB Neuroscience - KU Leuven, Belgium, in the summer of 2027 for a research internship. I am currently working on successor alignment with a collaborator and am actively seeking research opportunities in AI safety and mechanistic interpretability. I am also doing the Technical AI Safety course at BlueDot.

If I ask enough stupid questions, maybe one day I'll run out of reasons to call them stupid.