Piyush Ghai
Staff Applied Scientist at Relativity
San Francisco, California, United States
Summary
🤩
Rockstar🎓
Top SchoolPiyush Ghai is a machine learning engineer and applied scientist with 10+ years of experience building production-grade AI systems for privacy, compliance, and legal domains. Currently a Staff Applied Scientist at Relativity (and recently an MLE at Uber), he leads teams that bridge NLP research and enterprise deployment, delivering systems for PII detection, privilege classification, and reliable RAG pipelines. His background in distributed training at AWS—where he helped achieve fastest training times for BERT and scaled to thousands of GPUs—gives him deep expertise in making large models efficient and robust. An active open-source contributor, he’s improved model-serving and backward-compatibility tooling in projects like awslabs/multi-model-server and Apache MXNet, focusing on modularity and CI/CD resilience. He combines rigorous research instincts with a pragmatic engineering bent, often tackling hallucinatory behavior and explainability challenges to make LLMs trustworthy for mission-critical workflows.
10 years of coding experience
6 years of employment as a software developer
High School, High School at Delhi Public School - R. K. PuramStanford Lead, Stanford Lead at Stanford University Graduate School of BusinessBachelor's Degree B.E ( Information Technology ), Bachelor's Degree B.E ( Information Technology ) at Netaji Subhas Institute of TechnologyHigh School, High School at New Era Public School - IndiaMaster’s Degree Computer Science, Master’s Degree Computer Science at The Ohio State University
English, Hindi
Github Skills (12)
mxnet10
machine-learning10
deep-learning10
backward-compatibility10
python10
cicd10
scala9
onnx9
boto9
testing8
modeling8
trainings8
Programming languages (4)
Github contributions (5)
awslabs/multi-model-server
Jul 2018 - Oct 2018
Multi Model Server is a tool for serving neural net models for inference
Role in this project:
Back-end Developer
Back-end DeveloperContributions:64 commits, 44 PRs, 18 pushes in 3 months
Contributions summary:Piyush's commits focused on refactoring and improving the organization of the codebase. The primary activity was moving code related to model exporting into utility files to improve code cleanliness. These changes involved creating helper classes and methods related to ONNX model conversion and manifest generation. The impact of these changes was to improve the modularity of the model export tool.
pytorchmxnetservingdeep-learninginference
apache/mxnet
Jul 2018 - Jun 2019
Lightweight, Portable, Flexible Distributed/Mobile Deep Learning with Dynamic, Mutation-aware Dataflow Dep Scheduler; for Python, R, Julia, Scala, Go, Javascript and more
Role in this project:ML Engineer & DevOps Engineer
ML Engineer & DevOps EngineerContributions:29 commits, 34 PRs, 916 comments in 10 months
Contributions summary:Piyush primarily contributed to the Model Backwards Compatibility Checker (MBCC) within the MXNet repository. Their work included adding and refactoring model files, adding scripts to run older versions of MXNet models, and implementing the runtime function for MBCC files. They also set up the JenkinsFile for MBCC and implemented S3 uploads, demonstrating contributions to the CI/CD process. Furthermore, the user addressed test flakiness, added tolerance levels for regression checks, and added a Scala inference benchmark.
pythonschedulerdataflowmutationdata-science
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.