About Me

I recently earned my PhD in Electrical & Computer Engineering at Georgia Tech, where my research interests broadly revolved around characterizing learning and intelligence in artificial and biological systems. My thesis used RNNs as models for cortical circuits to understand structure-function relationships in the canonical cortical microcircuit under the purview of predictive coding, using a combination of tools from high-dimensional geometry, neuroscience, and signal processing. Additionally, it also developed methods to construct and train more biologically plausible deep learning models that are easy to scale and train with theoretical guarantees on learning efficiency and performance. Towards these ends, I spent my time studying both machine learning and theoretical + computational neuroscience, with a healthy smattering of many topics in applied mathematics. I was primarily advised by Dr. Hannah Choi and co-advised by Dr. Chris Rozell.

I am currently on the industrial (research scientist in AI/ML Interpretability, Alignment, Safety, and Theoretical + Computational Neuroscience) and academic (postdoc) job markets! Feel free to reach out if you’d like to chat about understanding computation in natural and artificial systems, as well as using these insights to develop principled, mathematically-grounded tools that can interpret and steer LLMs and/or foundation models.

Ongoing Research

My research thus far often leverages structure (geometrical and topological) in the representations and architectures of artificial & biological neural networks so as to render them more interpretable and thereby discover their governing principles. Some ideas that I actively think about in these contexts are:

• Sparse, low-rank, and low-dimensional approximations of data.
• The effects of regularization & architecture on representations learnt by neural networks.
• Disentangling of data manifolds, the quantification of representations that live on them, their evolution across space-time, and the changes that different perturbations bring about in them.
• Generalizability and potential for transfer of representations across different tasks and domains.

My long-term goals of understanding learning + intelligence combined with my facility for math & engineering have led to the following (somewhat) more tangible goals that I try to actively contribute to with my research:

• Development of mathematical models and computational methods that effectively characterize as well as elucidate changes in structure, organization, and information processing in deep neural networks and the brain across different stages of development.
• Development of “better” ML/AI systems that require less supervision and/or data, learn progressively, and are more amenable to changes in task structure (along with supporting foundational theory or guarantees whenever possible!)
• Tackling AI Alignment from first principles, i.e., trying to fundamentally ensure deep learning models as and when deployed, do what we want/intend for them to do (relatively recent foray!)

Recent-ish News

• May 2025: Our paper studying the architectural biases of the canoncial cortical microcircuit has been accepted at Neural Computation! This work was also highlighted as a talk at COSYNE ‘23 in Montréal!
• April 2025: I participated in the 2025 Frontiers in Science Conference and Symposium at Georgia Tech (focused on Intelligence + Neuroscience + AI), and won a best poster award!
• March 2025: I presented our work on constructing biologically-constrained RNNs that respect Dale’s law while incorporating sparse connectivity motifs at COSYNE 2025! The preprint of our paper is also available.
• February 2025: I am being supported by Open Philanthropy’s Career Development & Transition Funding through Spring ‘25 as I upskill in AI Safety and Alignment. Thanks, Open Phil!
• February 2025: I’ve successfully defended my PhD! My thesis is titled Through the RNN Looking Glass: Structure-Function Relationships in Cortical Circuits for Predictive Coding, and the dissertation document, defense talk, as well as slides are all accessible online!

Interests & Hobbies

Machine Learning: Unsupervised/self-supervised learning, dimensionality reduction & manifold learning, metric/similarity learning, and learning with structured sparsity.
Mathematics: Matrix & tensor decompositions, column subset selection, low-rank approximation, metric embeddings, convex geometry, optimization, group & representation theory, differential geometry & topology, and information geometry.
Neuroscience: Neural (i.e., population and sparse) coding, predictive coding, synaptic plasticity & learning rules, models of brain structure & organization, and connectomics.

Outside of academic and scientific pursuits, my hobbies include reading 📖, listening to + studying classical music 🎼, watching + playing racquet sports 🎾, trying (and often failing) to keep up with cool movies + TV shows 🎥, and solving every Rubik’s cube variant 🎲 I can get my hands on.

I am incredibly fond of cats 🐈, enjoy history of almost any kind 📜, still identify as an ardent Federer fan 💜, and remain a Bombay kid 🌏🏠👶 at heart for life. 🌈.

Social Network