Portrait of Shubhranshu Singh

Applied ML Researcher

Shubhranshu Singh

I am an Applied Researcher at eBay working on language models for buyer-seller conversational systems. My work bridges applied research and production ML, spanning data pipelines, model architecture iteration, training and evaluation frameworks, and deployment.

Before eBay, I was an AI Research Engineer at VOAIGE, where I worked on perception for robotics applications, including promptable segmentation and 3D geometry estimation.

I completed my M.S. in Electrical and Computer Engineering at Carnegie Mellon University, and interned at NVIDIA on autonomous vehicle perception datasets and weakly supervised segmentation.

I am broadly interested in computer vision, 3D perception, robotics, and NLP.

Education

Carnegie Mellon University logo

Carnegie Mellon University

M.S. Electrical and Computer Engineering - Applied Study

Indian Institute of Technology Gandhinagar logo

Indian Institute of Technology Gandhinagar

B.Tech. Electrical Engineering, Minor in Computer Science

Skills

Python PyTorch NumPy SQL C/C++ TensorFlow PySpark Pandas Open3D MMEngine Git AWS

Experience

eBay logo
Sep 2025 - Present

Applied Researcher, eBay

Evaluating large language models to improve buyer-seller chatbot interactions.

VOAIGE logo
Jan 2023 - Aug 2025

AI Research Engineer, VOAIGE

Improved promptable segmentation runtime by 10x, enabled object filtering through 3D geometry estimates, and led MMEngine pipeline development that reduced experimentation time by 66%.

NVIDIA logo
May 2022 - Aug 2022

Deep Learning Software and Research Intern, NVIDIA

Processed 550,000-image AV datasets with SQL and Python, developed ETL pipelines, and analyzed DiscoBox robustness under noisy bounding box supervision.

Caltech logo
May 2019 - Jul 2019

Summer Undergraduate Research Fellow, Caltech

Built a ResNet-based CNN to reverse atmospheric distortion in telescope imagery and improve faint-source detection while preserving flux measurements.

Selected Projects

Visual Grounding

Modified a detection transformer with cross-attention to identify image objects from text prompts, reaching 36.9% mAP.

Project Website

Few-shot Bioacoustic Event Detection

Implemented prototypical networks in NumPy from scratch and reached 90.5% accuracy with a logistic-regression-based network.

PDF

Publications and Patents

Patent Pending · NVIDIA Corp · 2024

Object Segmentation Using Machine Learning for Autonomous Systems and Applications

Alperen Degirmenci, Jiwoong Choi, Zhiding Yu, Ke Chen, Shubhranshu Singh, Yashar Asgarieh, Subhashree Radhakrishnan, James Skinner, and Jose Manuel Alvarez Lopez.

Patent
NeurIPS 2023 BUGS Poster

Adversarial Robustness Unhardening via Backdoor Attacks in Federated Learning

T. Kim, J. Li, N. Madaan, S. Singh, and C. Joe-Wong.

NeurIPS BUGS Poster
AISTATS 2023

Characterizing Internal Evasion Attacks in Federated Learning

T. Kim, S. Singh, N. Madaan, and C. Joe-Wong.

PDF
Software Impacts, 15, 100469 · Elsevier · 2023

pFedDef: Characterizing Evasion Attack Transferability in Federated Learning

T. Kim, S. Singh, N. Madaan, and C. Joe-Wong.

ScienceDirect
CrossFL 2022 · MLSys 2022

Grey-Box Defense for Personalized Federated Learning

T. Kim, S. Singh, N. Madaan, and C. Joe-Wong.

Best Poster Award
CrossFL Program
SPIN 2020

Neural Network Based Block-Level Detection of Same Quality Factor Double JPEG Compression

A. U. Deshpande, A. N. Harish, S. Singh, V. Verma, and N. Khanna.

IEEE