Computational Scientist · Scientific R&D
Modelling, Sensing & Scientific AI
I make hard scientific problems computable.
I've worked on plasma physics, optoelectronics, geoelectrical sensing, crop models, remote sensing and hydrology. The problems changed; computation stayed the common element — numerical modelling, differential equations, scientific software and data.
More recently AI has become another part of that toolbox, both for scientific computing and for evaluating frontier AI systems on hard scientific problems.
Currently
Apr 2025 →
Indian Institute of Science (IISc), Bengaluru
Earth-observation research on soil moisture and watershed hydrology in semi-arid Karnataka — satellite retrieval, field measurement and causal evaluation of watershed interventions.
Oct 2025 →
Chakradhari Computational Technologies Pvt. Ltd., Bengaluru
An independent R&D effort I co-founded. We take hard scientific problems and see what computation can do with them — currently sensing, inverse problems, Earth observation and scientific software.
May 2026 →
Mercor Intelligence
Scientific evaluation and refinement of frontier AI systems, including models before public release. I write hard scientific problems, find where models fail, and study whether they are reasoning through a problem or recognising it.
What I've worked on
I didn't plan this as one continuous research programme. I moved across problems that looked interesting and used computation on them.
UA Huntsville
Particle-in-cell simulation extended from 2D electrostatic to 3D electromagnetic, parallelised with MPI/OpenMP.
Alabama A&M · Paderborn
Laser–matter interaction experiments, then Fortran Maxwell–Bloch solvers for light propagation in excitonic systems.
UCLouvain
Coupled hydro-geophysical models linking root architecture, water uptake and electrical signature — field and laboratory ERT.
Forschungszentrum Jülich
Coupling process-based crop models with functional–structural root models across C/C++, Fortran and Python.
IISc Bengaluru
Sentinel-1 soil moisture retrieval, active–passive fusion with SMAP, and decade-scale watershed evaluation across semi-arid Karnataka.
Independent · Mercor
Writing hard scientific problems and evaluating how frontier models handle them — running alongside the hydrology, not after it.
The last two are happening at once. Watershed hydrology at IISc and frontier-AI evaluation are both current work — not consecutive chapters.
Research
Earth Observation & Hydrology
Satellite radar and passive microwave turned into soil moisture you can actually use — retrieval, sensor fusion and decade-scale records over semi-arid monsoon catchments.
Representative work — SAR-Based Soil Moisture Estimation with Machine Learning
5 projects
Environmental Sensing & Inverse Problems
Coupled hydro-geophysical models linking root architecture, water uptake and electrical signature, from virtual rhizotrons to field-scale resistivity surveys.
Representative work — Doctoral Thesis: Non-Invasive Geo-Electrical Imaging of Plant Roots
6 projects
Agricultural Modelling & Digital Twins
Making independently developed scientific codes talk to each other — process-based crop models coupled to functional-structural root models, packaged so others can run them.
Representative work — Agricultural Digital Twin Model Coupling
3 projects
Computational Physics
Particle-in-cell plasma simulation, Maxwell-Bloch solvers for light-matter interaction, and the laboratory optics that preceded both.
Representative work — Plasma Physics Simulations
3 projects
Scientific AI
Writing hard scientific problems, then testing how frontier AI systems handle them — including models before public release.
Representative work — Benchmarks that test whether a model reasons
1 project
Publications
Nordic Machine Intelligence, 6(1), 39–54.
in silico Plants, 8(1), diag002.
Vadose Zone Journal, 19(1), e20082.
Deprecated
Twenty years of learning to compute, written out as a version history. Each stage solved something the previous one couldn't, and each has since been superseded.
v1.0
2006C++ at BMSIT, Bengaluru. Sorting algorithms, microcontroller assembly, and MATLAB filters. Believed at the time that programming was technically inferior to designing circuits and antennas.
v2.0
2010Fortran and MPI at Huntsville. A 2D electrostatic particle-in-cell code extended to 3D electromagnetic, parallelised, and queued overnight. Submit, sleep, hope. A wrong sign in a boundary condition cost a day, so you learned to read your own code properly.
v3.0
2016Python at UCLouvain. Root architectures into meshes into finite-element electrical models — up to 500,000 tetrahedra — and back out as something you could compare against a field measurement.
v4.0
2023Jülich. Less about writing code than about making other people's codes talk to each other across C, C++, Fortran and Python, with timestep synchronisation and two-way state exchange. Then containerising it so the result would still run somewhere else.
v5.0
2025The loop writes itself now. What does not write itself is the decision about which problem is worth posing, whether the answer is actually right, and how you would know. That turns out to be most of the job.