Education
-
Imperial College London
BSc physics with theoretical physics, 2024–2027. First class.
-
Programming
Python, MATLAB, C++
-
Tools
PyTorch, TensorFlow, CUDA, PyMC, Cython, Numba, quantax, concurrent.futures, Slurm, JAX.
Honours
-
Talks at Caltech CMS and Caltech Seminar Day ’26
On multi-scale dynamics of accretion feedback with sub-grid black holes. PDF.
-
AI+Science Summer Conference ’26
Invited. Hosted by UChicago, SkAI, and Schmidt Sciences.
-
Talk at the Blackett Laboratory, Imperial College London ’25
On the promise of physics-informed machine learning. PDF.
-
John Locke Essay Competition ’23–’24
High commendation in philosophy (’23, ’24) and in theology (’23).
-
RCSU Science Challenge ’23
Finalist, invited to Imperial College London.
-
First place, IMC Algothon ’26
-
First place, DocSoc Marshall Wace ’26
-
First place, Bloomberg Data Challenge ’25
-
First place, G-Research London Quant Competition ’25
-
Second place, IC Hack ’25, ’26
-
Top 1%, IMC Trading Prosperity ’25
-
FIDE International Master, Online Arena
Experience
-
Research fellow, Caltech
September 2025 – present.
Anima AI + Science Lab, supervised by Anima Anandkumar. Capturing black hole accretion-driven feedback for galaxy-scale evolutions, learning chaotic dynamics on a sphere, and scaling neural quantum states for strongly correlated electronic structure. Open-source contributor to neuraloperator.
-
Research assistant, Virtus Solis
January 2024 – present.
Generative antenna design, and simulations for the Molniya orbit demonstration. Off-axis diffraction models, and a derivative-free beamforming optimisation algorithm using Lipschitzian methods.
-
Founding engineer, Hanomi
August 2025 – March 2026.
Machine learning and graph algorithms for computer-aided design.
-
Machine learning research, PhysicsX
May – August 2025.
A physics-informed neural operator for the Reynolds-averaged Navier–Stokes equations.
-
Systems team, Imperial College Aerial Vehicles
September 2024 – January 2025.
Twin-delayed deep deterministic policy gradient for dynamic motion planning and navigation.
-
Intern, Vantage Research
May 2023 – January 2024.
Unscented and extended Kalman filter and particle filter methods for joint state-parameter estimation in non-linear biophysical systems. Virtual population generation using a modified Metropolis–Hastings algorithm.
-
Intern, Indian Institute of Technology Madras
June – September 2022.
Charge carrier density, and the energy-level difference between the conducting band and the Fermi level, in n- and p-doped germanium and silicon. Supervised by Dr Bhattacharya.
Projects
-
Feb 2026
On the Bayesian characterisation of diffusivity through non-linear heat losses
Hierarchical Bayesian inference to determine the thermal diffusivity of aluminium via periodic temperature waves, treating local wave parameters as variable across frequencies while constraining diffusivity and heat loss as global constants. Hamiltonian Monte Carlo sampling and cross-validation show that including a linear heat-loss term improves predictive accuracy without being statistically significant, yielding a diffusivity of $9.6 \times 10^{-5}\ \text{m}^2\text{/s}$, in close agreement with the literature. GitHub, and a PDF.
-
Dec 2025
Beyond closure models: learning chaotic systems via physics-informed neural operators
An implementation of arXiv:2408.05177. GitHub.
-
Dec 2025
On the approximate Bayesian computation of background count using an Sr-90 source
A probabilistic method to estimate the background count rate using approximate Bayesian computation from a Strontium-90 source. Geant4-derived attenuation curves model the distance-dependent count rate over a range of 50 cm, yielding a posterior background estimate of $\kappa = 0.055 \pm 0.023\ \text{s}^{-1}$, consistent with the independently measured value of $0.060 \pm 0.025\ \text{s}^{-1}$. Posterior predictive checks show a conservative uncertainty budget, with 76% of data points within the $1\sigma$ credible interval. PDF.
-
Dec 2025
On the calibration of non-linear actuator motion using phase-shifting interferometry
Interferograms across a range of actuator speeds (5,000–35,000 µstep/second) compare phase-shifting calibration with crossing-points analysis. After accounting for measurement scatter with the Birge ratio, we recover weighted estimates $\lambda_{CP} = 531.08 \pm 0.22$ nm and $\lambda_{PSI} = 531.98 \pm 0.17$ nm. Against the nominal wavelength of $532 \pm 1$ nm, phase-shifting calibration is closer to the reference ($z \approx -0.02$) than crossing points ($z \approx -0.90$); a t-test ($t \approx -3.24$) confirms a statistically significant difference between the two methods. PDF.
-
Jun 2025
Physics-informed neural network for Grad–Shafranov equilibria
With Thomas Ashley, Arvind Chandrasekar, and Adam Skidelsky Gaathon. PDF.
-
2024
Mixture density network for time-series forecasting
Built for Imperial's and QRT's Algothon '24, for robust time-series forecasting over large datasets, and adapted for Jane Street's Real-Time Market Data Forecasting competition, 2024. With Conrad Trey Indradjaja and Oles Chaban
-
Nov 2023
On the expectation of real zeros
A formula for the expectation of real zeros of a random k-th order polynomial with Laplacian random variables as coefficients, derived as an integral and evaluated for small integer k up to 100. Approximation techniques give absolute upper and lower bounds and an asymptotic relationship, extended to other intervals on the real line. Collaboration with my twin. PDF.
-
Apr 2023
On the thermodynamic stability of 3d low-spin octahedral complexes
Studied experimental methods for determining $\log\beta$, crystal-field stabilisation energies, and the factors affecting complex stability, drawing on IUPAC surveys, Lange's Handbook, and NIST data across over 60 complexes with three trials each. Linear regression related ionic charge density to $\log\beta$, alongside analysis of valence-bond theory, orbital overlap, the Jahn–Teller theorem, the nephelauxetic effect, hard-acid–soft-base theory, inner-orbital hybridisation, π-bonding, the chelate effect, entropy, and heterocycles. PDF.
-
Mar 2023
Anharmonicity in the non-linear simple and double pendulum
Solved the simple pendulum via the elliptic integral $K(\kappa)$, expanded as an infinite series in $\sin(\phi_0)$ using the Fubini–Tonelli theorem, and derived a new closed form for the period in $\phi_0$. Performed phase-space stability analysis by linearising at critical points via the Jacobian, and used fourth-order Runge–Kutta to simulate the simple, double, and spherical pendulum in Python and C++, demonstrating chaotic behaviour, small-angle harmonic reduction and coupled oscillation, and the double pendulum's normal modes. PDF.
-
Mar 2023
Anharmonicity in the damped, non-linear pendulum
Studied anharmonic power in the simple pendulum and its effect on measuring g. Fourier analysis uncovered higher harmonics in the large-amplitude regime, with a discrete Fourier series written in Scilab, alongside exact numerical solutions and experimental data collected with the Vernier suite, analysed through nonlinear regression and parameter optimisation. Corrected the period for mass distribution — inertia, string flexibility and mass, buoyancy, and added mass via Bernoulli's equation for unsteady flow. PDF.
-
Jan 2023
On the variational Monte Carlo method
Variational Monte Carlo calculations for the harmonic oscillator, hydrogen atom, and helium atom, conclusively obtaining the ground-state energies for all systems studied. Guided by Pierre Guichard, Institut de Physique et de Chimie des Matériaux de Strasbourg. PDF.
-
Jan 2023
On the magneto-optic Faraday effect
Analysed Jones vectors and circular and elliptical polarisation, numerically modelling the magnetic field profile with a novel expression. Obtained the solenoid's constant and processed Faraday-rotation data to a Verdet constant of $169.54 \pm 32\%$ degrees/T·m at 310 K. PDF.
-
2022
Monte Carlo methods in statistical mechanics
Explored the Peierls argument and simulated the 2D Ising model with the Metropolis–Hastings algorithm, studying energy, magnetisation, specific heat, and susceptibility across temperature, and resolving the second-order phase transition at the Curie point. Supervised by Pierre Guichard, Institut de Physique et de Chimie des Matériaux de Strasbourg.
Separately, simulated the 2D XY model in C++ and Python using the Hoshen–Kopelman labelling algorithm, demonstrating the Kosterlitz–Thouless transition, and explored polynomial-time approximations for n-vector models. GitHub.
-
2022
The generalisation of Wilson's theorem to arbitrarily long prime gaps
Second author, with Ansh Pincha. Derived Wilson's theorem and Clement's theorem and generalised them to arbitrarily long prime gaps, giving a necessary and sufficient condition for $p$ and $p + 2\lambda$ to be consecutive primes for $\lambda \in \mathbb{Z}^+$. Used this to construct a prime 2-tuple counting function and its closed form, and tested its validity for the twin-prime case. PDF.
-
2021
On Crystal field theory
An experimental investigation of cobalt(II) and copper(II) complexes and the effect of charge density on crystal-field splitting energy, across aqua, cuprate, tetramminediaqua, hexammine, and EDTA complexes. Used UV-Vis spectrophotometry to obtain absorbance spectra, and WolframAlpha-based computational modelling to show the axial orbital overlap of $d_{z^2}$ with the electron pair perpendicular to the xy-plane in the elevated, non-degenerate $e_g$ orbital, with a brief look at the Jahn–Teller effect.
Previously involved in
-
The Notes Repository
Free, structured undergraduate physics and mathematics for independent learners.
I’ve spent a couple years through high school bumping elbows with “higher material” like Bayesian inference, Jacobians, Lagrangians, Markov chains, and Fourier transforms. They’ve always had a fearsome reputation for difficulty. The theories are beautiful, but quite difficult to learn if you are not already well trained in advanced mathematics and physics. There are excellent textbooks out there but I found that reading such material is a dismal and trying experience if you do not know what to consult. Typically the first few pages fly through math, sometimes presenting an entire semester’s worth of content in a few brief chapters. It is often difficult to find your way around which material to read and in what order to read it. Most texts are built for very specific purposes directed at a specific group of people with specific requisites. Lacking knowledge of it impedes learning. The Notes Repository is built to address this need. This is not the place to be if you design military-grade radars or are working on the next Apollo mission. It’s for the curious, independent learner. Here, the intention is to establish a solid foundation in undergraduate mathematics and physics in a structured, directed way. In the future, I intend on developing all the material from the ground up but there are many people on the internet who are far more qualified than I am — it is finding the right content that is time-consuming. You’d be surprised by how much can be learned for free. Currently, the content hosted will largely revolve around Physics and Mathematics — I hope to extend this to Computer Science as well. This is a realization of my effort for open-education.
-
CarbonZero
Modelling the viability and lifetime cost of carbon solutions.
Rigorously determines the viability of carbon solutions through mathematical modelling of net operational costs and monetary benefits, accounting for population growth, adoption rate, learning rate, and projected demand, with scenarios drawn from IPCC and IEA reports. Developed methods to estimate $\text{CO}_2\text{e}$ emissions for industrial projects and the lifetime cost of the carbon credits needed to offset them, and has coordinated with over a dozen non-profits to verify their impact. Currently developing alternative climate solutions for the general public. crbnzero.com.
-
Project Sicilian
Chess as a vehicle for socio-economic equalisation.
Co-founded with my twin, Ansh Pincha: a non-profit teaching chess to children from underprivileged communities, on the premise that critical thinking and independence — not one-off aid — are what break the poverty cycle. Chess, in effect, as a gymnasium for the mind.
-
Green Origin
Study guides written and donated to six state-board schools in Tamil Nadu.
Founded as a non-profit during the pandemic; led the team to the Most Impactful Award at the Global Goals competition. Established an Interact Club under the Rotary, ran crowdfunding campaigns against hunger, and worked with companies on their corporate social responsibility. Co-founded an inter-continental resource-sharing network for International Baccalaureate students.
Co-authored (Physics; co-author, Chemistry) a set of study guides for 10th-grade State Board students — around 150 hours of work across four subjects, checked for accuracy by school faculty — and donated them to six schools in Tamil Nadu, the largest being Sevalaya. Addressed the UN Sustainable Development Goals for zero hunger, good health and wellbeing, quality education, industry, innovation, infrastructure, sustainable cities, and partnerships for the goals.
-
Project Wren
Journalism on what goes right.
A polemic against the negative bent of mass media, and an antidote to despondency: honest, comprehensive journalism on what goes right in the world. Began after noticing that the news was punctuated with everything wrong; has since hosted more than fifteen writers and editors, drawn 40,000+ views, and reached 200+ students through outreach, publishing 140+ stories. projectwren.org.
Languages
English, Hindi, French, and Tamil.