Vaibhavi Singh

I'm an MS student in Computer Science at NYU Courant, specializing in machine learning with a focus on multimodal models & representation learning. I'm particularly interested in understanding how vision-language models reason & how to train effective representations with limited data.

Currently, I'm conducting research on reasoning in multimodal LLMs, & investigating self-supervised learning in low-resource regimes.

Previously, I worked as an ML Engineer building clinical risk prediction models, & spent several years at Adobe & Salesforce optimizing ML systems & large-scale graphics pipelines.

Email  /  LinkedIn  /  GitHub  /  Twitter  /  Writings

Research

I'm interested in agentic reasoning & representation learning. My current work focuses on evaluating reasoning capabilities in vision-language models & efficient self-supervised learning under resource constraints.

Evaluating the limitations of current multimodal models on structured reasoning tasks, with a focus on evaluation challenges & qualitative failure analysis.

Investigating self-supervised representation learning under severe data, resolution (96×96), & compute constraints by training DINOv1 from scratch. Observed strong performance on CUB-200, miniImageNet, & SUN397, & explored the roles of domain-aligned data, tokenization density, & evaluation methodology in constrained regimes.

Academic Service

Ethics Reviewer — Datasets & Benchmarks Track
Technical Reviewer — Workshops - (UniReps) Unifying Representations in Neural Models, (ML4PS) Machine Learning and the Physical Sciences

Skills

Experience

Built clinical risk prediction models (XGBoost, TCN) achieving 0.87 F1-score through feature engineering, SMOTE for class imbalance, & hyperparameter optimization. Processed sparse EMR data for early-stage healthcare applications.

Optimized petabyte-scale data ingestion pipelines, reducing latency by 30% for Einstein AI & real-time analytics. Scaled multi-tenant Kubernetes infrastructure on AWS for 200+ microservices.

Implemented heterogeneous compute (CPU/GPU) optimizations in graphics libraries for Photoshop Neural Filters, enabling real-time HDR style transfer on Intel & Apple Silicon for 20M+ users. Developed graphics shaders for the Vulkan rendering backend for Substance 3D in Color Engine, enabling high-fidelity 3D workflows on Linux & macOS. Implemented multithreaded parallel processing in media libraries powering Premiere Pro & led performance profiling efforts to enforce strict latency SLAs across the Creative Cloud suite.

Optimized high-throughput Java microservices on Adobe Cloud Platform, reducing build times by 25% & infrastructure costs by 12% for 10M+ DAU through performance profiling & resource tuning.

Education

Research focus: agentic reasoning & representation learning

Coursework: Deep Learning (Yann LeCun), Natural Language Processing (Eunsol Choi), Computer Vision (Saining Xie)

Graduated in the top 10% of the department
Recipient of EPFL-Swiss Government scholarship (Scala Days 2019)
Google Summer of Code Mentor, Anita Borg Institute

Last updated: January 30, 2026
Website template from Jon Barron