Neham Jain bio photo

Email

LinkedIn

Github

Google Scholar

Professional Experience


Meta Reality Labs

Research Engineer | Pittsburgh, PA | June 2025 - Present

Data Pipelines for Training Embodied AI Agents

  • Embody3D dataset release: Built QA tools, coordinated annotation of 1.6M videos, and curated 500 hours of high-quality data — 187 GitHub stars, 3 citations, adopted by internal teams for training.
  • Built automated quality filtering algorithms to eliminate annotation bottlenecks and ensure high-quality data for training.
  • Achieved 75% reduction in tracking error rate by designing a self-supervised data generation pipeline to train keypoint model.

Adobe Research

Research Engineer Intern | San Francisco, CA | Summer 2024

Scalable Pipeline for Training 3D Foundation Models

  • Created a pipeline to convert synthetic mesh assets into relightable Gaussian Splats for training 3D Foundation Models.
  • Wrote custom CUDA kernels to replace spherical harmonics with neural features for relightability and faster training.
  • Scaled the training pipeline to 32 nodes and improved training time from 6 minutes per asset to 30 seconds.

Adobe Research

Research Scientist Intern | Bangalore, India | Summer 2022

Counterfactual Explanations of Visual Recommender Systems [Paper]

  • Worked on counterfactual explanations for recommendation systems based on vision language models.
  • Computed the minimal meaningful perturbation to an item’s image-embedding that would remove it from a user’s recommended list. Used CLIP to connect image features to textual labels to lend meaning to the perturbations.
  • Work was accepted at WWW 2024 and a patent has been filed at USPTO.

Subex AI Labs

Machine Learning Intern | Bangalore, India | Summer 2021

Information Extraction from Unstructured Invoices

  • Developed an end-to-end pipeline for multilingual invoice understanding and table extraction using object detection.
  • Fine-tuned a multimodal document transformer model (LayoutLMv2) to extract form fields using token classification.


Education


Carnegie Mellon University

Master of Science in Robotics | May 2025

  • Advisor: Prof. Ioannis Gkioulekas
  • GPA: 4.11/4.0
  • Relevant Coursework: Deep Learning Systems, Learning for 3D Vision, Robot Learning, Physics-based Rendering

Indian Institute of Technology (IIT) Madras

Bachelor of Technology in Electrical Engineering | June 2023

  • CGPA: 9.24/10
  • Minor: Machine Learning
  • Relevant Coursework: Modern Computer Vision, Machine Learning, Computational Photography