California Institute of Technology
2017 – 2023PhD in Computational Neuroscience
Pasadena, California
- Thesis: Engineering flexible machine learning systems inspired by human intelligence
- Advisor: Dr. Matt Thomson
I build machine learning systems inspired by how human intelligence actually works — flexible, continually learning, and able to acquire new skills without forgetting the old ones.
I'm currently Head of AI at Workfabric AI, where I lead research on compounding intelligence: inferring feedback implicitly from a knowledge worker's sequence of human–computer interactions, and using it to post-train models that behave as digital twins of individual enterprise workers.
Before this, I was the Founder & Lead Research Scientist at Legion Intelligence (previously Yurts AI), where I led research on domain-adaptable retrieval-augmented generation and on compressing LLM decoders for fully air-gapped deployments — using functionally invariant paths, the differential-geometry framework I developed during my PhD.
I completed my PhD at Caltech in computational neuroscience with Dr. Matt Thomson, a research masters at Carnegie Mellon with Dr. Ge Yang, and a B.Tech at IIT Madras.
My interests have been broad: accidental beginnings in biology, a niche in computational biology, converging on computational neuroscience, and most recently artificial intelligence inspired by biological systems. Along the way I've been fortunate to work with and be guided by some of the best minds in these fields.
PhD in Computational Neuroscience
Pasadena, California
Master of Science in Biomedical Engineering (research masters)
Pittsburgh, Pennsylvania
Bachelor of Technology, Biotechnology
Chennai, India
Head of AI
Palo Alto, California · Seed stage
Founder & Lead Research Scientist
San Francisco, California · Seed to Series B
A book on consciousness written with philosophers and physicists, my doctoral thesis, technical reports and articles from building retrieval and fine-tuning systems, and the “Why” series — essays where I chase a question from physics, mathematics or neuroscience until it explains something about how learning systems work.
During my PhD, alongside my work on bio-inspired AI, I took a keen interest in ancient Indian philosophy — particularly the theories of consciousness set out in Sanskrit texts. Because so much of that material is encoded in Sanskrit poetry, I worked with a group of researchers — philosophers and physicists who share the interest — on a manuscript that draws those theories out and places them alongside modern experimental work in neuroscience, offering viable Indian philosophical frameworks for building a larger account of the mystery we call consciousness.
With frontier LLMs advancing as quickly as they are, I'd like to bring these frameworks to bear on a question we will soon have to answer seriously: if and when we might declare an AI to be conscious.
My chapters “The Opposing Viewpoint” (ch. 2) and “Theory of Mind, Mirror Neurons, Empathy and Autism” (ch. 11).
Contributors Caltech, Rutgers, UCL, Temple University, UT Houston, New Children’s Hospital and BIHS.
Degree PhD, California Institute of Technology, 2023 · advised by Dr. Matt Thomson.
My dissertation asks what artificial learning systems can borrow from the brain, and makes the borrowing concrete. The first half treats a neural network’s weight space as a curved Riemannian manifold and samples functionally invariant paths through it, letting a network pick up new tasks, sparsify itself or gain adversarial robustness without destroying what it had already learned. The second half grows a layered network from a single cell: taking a cue from how the developing retina wires itself to the lateral geniculate nucleus, a spontaneous activity wave in the first layer plus a local learning rule in the second produces networks that match hand-designed ones on digit and gesture recognition.
Written at Legion Intelligence (formerly Yurts AI).
Published on Substack.