Experience
Research timeline
-
PhD Student
Formalising identifiability in settings that provide no ground-truth factors to compare against, and testing the framework on the interpretability of language models. The work covers the equivalence class an agent's affordances induce, whether identifiable solutions are reachable under finite data and a given optimiser, and what current evaluation metrics can and cannot certify.
advised by Dhanya Sridhar
identifiability · causality · interpretability · LLMs
-
Research Intern
Adapted language models for text-based reinforcement learning environments (Jericho), where the agent parses natural language observations, maintains a belief state, and acts in a combinatorially large action space. Model capacity was scarce and architectural choices were consequential. The question it left me with is the one I still work on: what separates a model that tracks a useful latent state from one that correlates with low-level observable features?
reinforcement learning · language grounding
-
Master's Student
Studied how acting shapes what is learnable: whether model-based agents acquire causal structure about physical mechanisms through interaction with their environment. Built controlled physics simulations to evaluate planning and long-term memory under distribution shift and novel interventions, in partially observable and stochastic environments. This is where affordance-based representation learning became the question — what a system can do and observe as a result, and whether the variables it learns reflect that structure of control.
advised by Devon Hjelm
agency · reinforcement learning · causal structure
-
Research Programmer
Empirical Inference, Max Planck Institute for Intelligent Systems
Developed real-time control and logging software for robotic manipulators, and was the primary developer of the accompanying simulation package. Research on self-supervised representation learning, with a focus on systematic generalisation. This raised the question I have worked on since: how to learn a representation that captures relevant structure in data, how to tell whether it has, and what relevant means.
with Bernhard Schölkopf · Stefan Bauer
robotics · self-supervised learning · open-source software · systematic generalisation
-
Bachelor's in Electrical Engineering
Indian Institute of Technology Kanpur
Research across convex optimisation, stochastic processes, robotic manipulation, and formal methods. Projects included physics simulations in C++ for deformable object manipulation, dynamic programs for supply-chain decisions, online convex optimisation, stochastic geometry and spatial point processes, and using Z3 to formalise reasoning in multi-agent social deduction.
with Dmitry Berenson (deformable object manipulation, University of Michigan) · Amar Sapra (supply chain, IIM Bangalore) · Indranil Saha (Z3 multi-agent reasoning, IIT Kanpur)
optimisation · physics simulation · stochastic processes · formal methods
Tools
Python · PyTorch · JAX · C++ · TypeScript · Rust · Git · Docker · LaTeX · Weights & Biases · HPC / SLURM