cv

Deputy Contingent Leader

Inter-IIT Tech Meet 14.0

Lead the group of 90 students representing IIT Gandhinagar in the 14th Inter IIT Tech Meet conducted at IIT Patna.

• Under my leadership, IIT Gandhinagar delivered its best performance to date with 4 medals
• Including the college's first gold and my individual bronze in a no-prep problem statement

Machine Learning Research Intern

Magikkraft, IIM Ahmedabad Incubatee

Worked on a CV and ML-based construction project tracking system that analyzes photos and drawings.

• Prepared a CV model for the detection of steel pipes on an Oil & Gas Plant Construction Plant using EfficientDeT and YOLO Architecture
• Prepared a Vision Model for Comparing 2D Engineering Drawing with site photos
• Explored different architectures for Object Detection & Counting, and working on models for interpreting CAD files

Research Intern - Dynamic Graph Coarsening for GNN Preservation

MISN Lab, IIT Delhi

Developed Mathematical Guarantees on a novel Dynamic Graph Coarsening Framework.

• Analysed theoretical Gromov-Wasserstein distance and gave novel RSA constant-based bounds
• Currently working on developing and testing a novel algorithm for Continuous Dynamic Graph Coarsening
• Under Review at JMLR

Management Coordinator

Technical Council, Indian Institute of Technology, Gandhinagar

Responsible for smooth execution and organisation of technical events undertaken at IIT Gandhinagar.

• Aiming to empower students through technical skills and foster a technical culture among the student community

Research Intern - Data Generation for Physics Informed ML modelling for PCB Simulation

IIT Delhi

Developed an indigenous PCB design and simulation software with Generative AI capabilities.

• Created a cross-validated benchmark thermal, elasticity, and electromagnetism simulation datasets for physics-informed machine learning
• Gained experience in FEM pipelines, mesh generation, and validation using MFEM, OpenEMS, and Elmer

Instructor, SC-392: Introduction to Machine Learning

IIT Gandhinagar

Co-designed and delivered a 7-lecture introductory ML course for 200+ participants from B.Tech, M.Tech, MA, M.Sc., and PhD programs.

• Led lectures and discussions on core ML concepts and supervised labs
• 120+ first-year undergraduates actively engaging from the first session

B.Tech

Indian Institute of Technology, Gandhinagar, India

Computer Science & Engineering

• Machine Learning
• Data Centric Computing
• Calculus of Several Variables
• Discrete Mathematics
• Probability Statistics & Data Visualisation
• Data Structures & Algorithms
• Theory of Computing

Class XII

Divine Life International School (GSHSEB)

Maths, Physics, Chemistry, CS

Class X

Devasya Intl. Public School (GSHSEB)

Summer Undergraduate Research Fellowship at Caltech

Caltech

Received fellowship for summer 2026

Best Student Paper Award

International Conference on Advanced Scientific Computing & Machine Learning

Received Best Student Paper award at ASCML 2026

Academic Citation for Academic Excellence

IIT Gandhinagar

Academic Excellence in Semester I of the academic year 2025-26

Bronze Medal - Inter-IIT Tech Meet 14.0

Inter-IIT Tech Meet

Bronze medal in AI/ML Category proposed by Jilo Health

JEE Advanced - AIR 1712

IIT

Secured All India Rank 1712 among 180,372 Students Appeared

JEE Mains - AIR 1364

NTA

Secured All India Rank 1364 among 1,113,325 Students Appeared

The Pareto Frontier of Stable Adaptation: Hard Switching Constraints in Online Convex Optimization

COLT 2026: The 39th Annual Conference on Learning Theory

Under review

Kernel Invariant Risk Minimization: Fixed-Point Causal Discovery in High Dimensions

COLT 2026: The 39th Annual Conference on Learning Theory

Under review

Defying Asymptotic Feature Dynamics in Heterophilic GNNs using Drichlet Oscillations

COLT 2026: The 39th Annual Conference on Learning Theory

Under review

Spectral Geometry and Minimax Optimality of Counterfactual Mean Embeddings under Partial Identifiability

COLT 2026: The 39th Annual Conference on Learning Theory

Under review

Multi-Stage Physics-Informed Neural Network for Jet Flame Emission Modelling

Advanced Scientific Computing & Machine Learning (ASCML) 2026, BITS Pilani K K Birla Goa Campus

ACCEPTED - Oral Presentation

Multi-Diagnostic Ensemble for Physics-Informed Neural Networks

Advanced Scientific Computing & Machine Learning (ASCML) 2026, BITS Pilani K K Birla Goa Campus

ACCEPTED

Capturing High Frequency Features using HF-Boost based Spectrally Adaptive Physics-Informed Neural Networks

Advanced Scientific Computing & Machine Learning (ASCML) 2026, BITS Pilani K K Birla Goa Campus

ACCEPTED

Efficient Temporal Hypergraph Coarsening Framework with Theoretical Guarantees

ICML 2026: The Forty-Third International Conference on Machine Learning

Under review at ICML 2026

Efficient and Reliable Dynamic Graph Coarsening with Theoretical Spectral Bounds

Journal of Machine Learning Research (JMLR)

Manoj Kumar*, Abhishek Gupta*, Ridham Patel, Sandeep Kumar - Under review

Dynamic Hypergraph Coarsening with structural properties & HGNN performance preservation

Investigating Coarsening methods for Temporal Hypergraph for various machine learning applications, preserving structural properties like cut, flow, conductance and Hypergraph Neural Network performance.

• Developed a novel framework for making Temporal Updates to Coarse Hypergraphs
• Achieved up to 55x speedup from a full-recomputation-based baseline
• Developed mathematical guarantees for deviation from baseline
• Under Review at AISTATS 2026

Physics Informed Diffusion Model for Flood Modelling and Prediction

Developing a first-of-its-kind diffusion model for flood prediction in both fluvial and pluvial contexts.

• Developed a preliminary version of the Graph-based Diffusion architecture for flood modelling in urban context
• Learned Diffusion (DDPM) and Graph Neural Networks Based Methods for Urban Flood Prediction

ProGNN: Residue-level GNN for Secondary Structure & RSA Prediction

Converted PDB/mmCIF structures into residue graphs for per-residue SS and ASA labels.

• Engineered node/edge features and built a PyG pipeline for dataset discovery and preprocessing
• Implemented a 2-layer GCN with dual heads for 3-class secondary-structure classification and RSA regression

Kernel Based Graph Condensation Algorithm for comparable GNN performance

Developed a graph distillation technique to reduce computational complexity, preserving GNN performance.

• Developed a scalable algorithm utilizing GNTK with Nyström approximation and kernel K-Means
• Enabled up to 80% graph size reduction with less than 1% GNN accuracy loss on Cora

Hub Based Graph Construction Techniques for HNSW

Proposed and implemented a supervised Graph Construction technique for the Hierarchical Navigable Small World (HNSW) algorithm for ANN search using Hubness theory.

• Explored the phenomenon of hubness in high-dimensional spaces
• Achieved perfect recall with 125x faster build times, slightly lesser query time & 82% fewer edges than vanilla HNSW

Graph Transformer Networks for Molecular Dipole Prediction

Developed a Graph Transformer Network-based model for predicting molecular dipoles using spatial positions of atoms in a molecule.

• Integrated 3D spatial positional encodings and attention-based message passing
• Model performance was compared with the graph neural network baseline

Next Line Predictor Chatbot with Multilayer Perceptron

Developed a Next-line predictor using a multilayer perceptron and deployed it using Streamlit.

• Trained a multilayer perceptron for various sets of hyperparameters and analyzed its impact on model performance