About
I recently completed my PhD at École Polytechnique, Paris, in the GeometriX group, advised by Maks Ovsjanikov. My work spans neural surface representations, 3D reconstruction, and shape correspondence. Prior to this, I earned a Master's in AI & Visual Computing from Ecole Polytechnique and an Electronics engineering degree from BITS Pilani, Goa.
Actively seeking research/engineering roles in domains where Vision and Graphics meet language.
Research Publications
Deformation Recovery: Localized Learning for Detail-Preserving Deformations
LJN is lightweight method for detail-preserving shape deformations using local Jacobians, where each triangle is considered as training example instead of the entire shape. LJN is thus data-friendly and can learn high-quality deformation (human) from as few as 60 pairs of shapes.
Presented at: SIGGRAPH-Asia, Tokyo, 2024
Physical Property Estimation and Optimization via Constrained Latent Space Exploration
A 3D Generative approach that estimates and improves physical properties while ensuring geometric plausibility by using latent space sampling within convex polytopes. Introduces a new dataset of annotated bottles.
Under-Review, 2024.
Self-Supervised Dual Contouring
SDC introduces self-supervised dual contouring for isosurface extraction, replacing supervised training with novel losses enforcing mesh-SDF consistency. It improves mesh extraction from SDFs that are produced by Deep Networks.
Presented at: CVPR (Spotlight), Seattle, 2024
Reduced representation of deformation fields for effective non-rigid shape matching
Marrying mesh-free approximation method with MLPs for non-rigid shape correspondence. Learning reduced deformation parameters to reconstruct smooth deformation, our approach enables efficient, limited supervision and achieves state-of-the-art results on shape matching benchmarks.
Presented at: NeurIPS, New Orleans, 2022
Implicit field supervision for robust non-rigid shape matching.
An auto-decoder framework which learns continuous deformation fields for shape deformation, using Signed Distance Functions (SDFs) as regularization. Achieves strong performance on noisy, real-world data despite training on clean meshes.
Presented at: ECCV (Oral), Tel-Aviv, 2022
Tracking pedestrian heads in dense crowd.
We introduce CroHD, a large annotated dataset for tracking in dense crowds, along with IDEucl metric to evaluate identity preservation. We also propose HeadHunter, a head detector combined with particle-filter-based tracking framework, which acheives superior results compared to state-of-the-art pedestrian trackers.
Presented at: CVPR, (Virtual), 2021
Solving Inverse Computational Imaging Problems using Deep Pixel-level Prior.
Uses autoregressive models as flexible signal priors for inverse imaging problems, enabling better reconstruction of images with texture.
Accepted at: Transactions on Computational Imaging, 2018
Projects
(demo coming soon)
Physically-Aware 3D Assets: Segment-driven Material Synthesis
This project explores segment-level optimal physically-based rendering (PBR) materials to create realistic 3D assets. It focuses on enriching 3D geometry with accurate material semantics, enabling high-fidelity rendering of object parts. Segmentations are derived automatically, and material types are inferred using language-driven queries, enabling realistic and interpretable rendering of objects.
(demo coming soon)
LLM + RAG Email Generation with Adaptive Templates
A template prompt (tone and structure rules) and a retrieved subset of client-based case files (RAG over a corpus) are fused in an LLM call to draft the email. The corpus is embedded and searched to select only relevant chunks before generation. Feedback text then updates the template automatically so later drafts improve without manual edits. An evaluation pass flags hallucinations and style issues and feeds a short report back into the loop. Outputs are the drafted email and the revised prompt.
Education
PhD: Analysis of Large-Scale 3D Shape Collections with Learning-Based Approaches
This thesis develops data-driven pipelines for shape deformation, correspondence, registration and optimization with limited supervision, supported by respective real-world applications.
École Polytechnique, 2025
Master’s: Implicit Latent Space Exploration for Shape Optimisation and Correspondence
École Polytechnique, 2020
Open-Source Contributions
Theano (Google Summer of Code, 2016)
- Built a new “GraphToGPU” optimizer, enabling 2–3x faster compilation of deep nets like ResNet50.
- Refactored computation graph using CGT-style grouping to reduce CPU-GPU memory transfers.
- Implemented Spatial Pyramid and ROI Pooling layers with C++ and CUDA backends.
Protein Geometry Database (Google Summer of Code, 2015)
- Extended PostgreSQL schema and Django ORM logic to support occupancy and deposition filtering.
- Redesigned user and search systems with Django auth and AngularJS, enabling tagged saved-searches.
Service
- Reviewer: NeurIPS 2025, IEEE Transactions on Visualization and Computer Graphics (TVCG)
Talks
- ECCV 2022, Tel-Aviv, Israel: Implicit field supervision for robust non-rigid shape matching.
- SIGGRAPH Asia 2024, Tokyo, Japan: Deformation Recovery: Localized Learning for Detail-Preserving Deformations.
Teaching
- Computer Graphics - CSE-306, Ecole Polytechnique
- Geometric Deep Learning - INF-631, Ecole Polytechnique
- Computer Animation - INF-633, Ecole Polytechnique
- Introduction to Computer Science/Programming - INF-361, Ecole Polytechnique