Structural engineer bridging mechanics and machine learning — building Physics-Informed Neural Networks for damage detection & structural optimization.
M.Tech in Structural Engineering at IIT Indore, beginning a Ph.D. at IIT Delhi. I work where structural dynamics, health monitoring, and deep learning meet — turning limited sensor data into reliable insight about how structures behave and fail.
I'm a structural engineer with a deep interest in computational mechanics and machine learning. My research centers on Physics-Informed Neural Networks (PINNs) — embedding the laws of structural mechanics directly into neural networks to detect damage and optimize designs without relying on large training datasets.
I completed my B.Tech at NIT Patna and my M.Tech in Structural Engineering at IIT Indore, and I'm now beginning a Ph.D. at IIT Delhi. Along the way I've contributed to the Delhi Metro Phase IV design, taught graduate and undergraduate mechanics courses, and authored research on differentiable truss optimization.
My toolkit spans classical structural analysis — Staad.Pro, ETABS, ANSYS — and modern ML frameworks like PyTorch, JAX, and Flax. I'm drawn to problems where rigorous physics and data-driven methods reinforce each other.
Advanced Solid Mechanics (Masters) & Design of Steel Structures (Bachelors). Guided students through numerical problems, prepared assignments, and ran doubt-clearing sessions.
Part of the team designing and analysing Delhi Metro Phase IV. Gained exposure to viaducts, bridge superstructures, and substructures.
Flexible pavement design — traffic loading, material selection, and layer thickness design using IITPave; prepared the Job Mix Formula.
A Physics-Informed Neural Network for damage detection in truss structures from limited-sensor data, benchmarked against Modal Strain Energy and Flexibility Matrix methods paired with TLBO and Democratic Particle-Swarm Optimisation.
A PINN-based optimization method to minimize truss weight under constraints on displacement, stress, and cross-sectional areas — gradient-based and entirely data-free.
Simulated the dynamic response of a 3D truss tower in Python under El Centro ground motion, capturing modal behavior and seismic response.
Trained a Random Forest classifier on healthy and known-damage vibration datasets to detect damage in a beam from vibration data alone.
Compared the seismic response of a building across different shear wall configurations to evaluate their effect on dynamic performance.
Computed and compared storey shear, storey drift, and mode shapes for a G+6 building using two seismic design methods in Staad.Pro.
A physics-informed, differentiable framework for truss size optimization in which a neural network predicts member cross-sectional areas while enforcing structural constraints through a mechanics-based loss function — enabling efficient gradient-based optimization without any training data.