Yash Vanjani

Software Engineer

San Francisco, California, United States7 yrs 2 mos experience

Highly StableAI Enabled

Key Highlights

Expert in Large Language Models and Machine Learning.
Proven track record in optimizing distributed systems.
Strong background in C++ and software development.

Stackforce AI infers this person is a Machine Learning and Distributed Systems expert in the SaaS industry.

Contact

Skills

Core Skills

Large Language Models (llm)Machine LearningDistributed SystemsC++Node.jsReact.js

Other Skills

JAXTransformersPythonDistributed File SystemsMultithreadingDistributed ComputingStorage VirtualizationDeep LearningProtocol BuffersObject-Oriented Programming (OOP)Concurrent ProgrammingFile SystemsApache ZooKeeperEnterprise StorageDesign Patterns

About

Yash graduated from Carnegie Mellon University in Dec 2022, specializing in Machine Learning & Systems. Presently, he works at Essential AI in the LLM Pre-training Infrastructure team. Before that, he worked at Nutanix in the Core Data Path - Stargate team, the core Distributed Storage service at Nutanix. Before CMU, he worked for 3 years as a Software Development Engineer 2 at Adobe, primarily working in C++, where he delivered multiple major features into production and owned several components of Adobe FrameMaker, a desktop-based word processor. Before that, he completed his undergrad at IIT Guwahati, majoring in Computer Science & Mathematics.

Experience

7 yrs 2 mos

Total Experience

1 yr 9 mos

Average Tenure

1 yr 9 mos

Current Experience

Essential ai

Member of Technical Staff

Sep 2024 – Present · 1 yr 9 mos · San Francisco, California, United States · On-site

Building open source foundation models.
Focus areas:
LLM training performance optimization using hybrid parallelism strategies like DP, FSDP, TP, SP, etc.
Efficient implementation of SOTA optimizers for AMD GPUs in MaxText - a JAX-based training framework.
Novel attention techniques in Fused Attention kernels.
Large-scale inference system for synthetic data curation.
Web-scale data pipelines.

JAXTransformersLarge Language Models (LLM)Machine LearningPythonDistributed Systems

Nutanix

2 roles

Member of Technical Staff 3

Feb 2023 – Aug 2024 · 1 yr 6 mos · San Jose, California, United States · On-site

Core Data Path - Stargate
Spearheaded the detection of disk hung issue in large HDDs, developed a solution utilizing filesystem/ext4, kernel, and internal indicators, leading to early detection and eliminating downtime of 24-48hrs under some conditions.
Designed on-demand LSM Memtable’s Checkpointing, enabling efficient crash recovery and data persistence.
Identified and fixed race conditions in core Distributed Storage service, decreasing crash-rate by 10%.
Created an LLM based bot running on an on-device Ollama server to summarize JIRA updates for a user by employing Retrieval Augmented Generation (RAG), enhancing employee work tracking.
Developed a granular product versioning system, streamlining feature management and simplifying back-porting.

Distributed File SystemsMultithreadingDistributed ComputingStorage VirtualizationC++Deep Learning+8

Member Of Technical Staff Intern

May 2022 – Aug 2022 · 3 mos · San Jose, California, United States

Core Data Path - Stargate
Developed a versioning system to enable granular level state management of features.
Designed Protocol Buffer schema for developers to add new features to the system, along with the functionality to register callbacks to identify state changes of features. Used Apache ZooKeeper for persistent storage of state.

C++Protocol BuffersDistributed SystemsObject-Oriented Programming (OOP)Apache ZooKeeper

Adobe

3 roles

Software Development Engineer 2

Promoted

Feb 2020 – Aug 2021 · 1 yr 6 mos · Delhi, India · On-site

Owned the software development of an enterprise server product, FrameMaker Publishing Server (FMPS) after a short span of 2 years at Adobe.
Integrated Adobe Experience Manager’s file versioning system to enhance the publishing functionality of the server.
Fixed memory leaks and increased server availability by 10x by improving the middleware design.
Added logging functionality in vertical scaling environment which eased the debugging process.
Built a crypto-system for storing enterprise data along with user credentials.
Developed APIs to support integration into Adobe Experience Manager.

Node.jsDesign PatternsExpress.jsMongoDBGitHubC+++3

Software Development Engineer

Jul 2018 – Jan 2020 · 1 yr 6 mos · Delhi, India · On-site

Was part of the development team of FrameMaker (FM), a desktop-based document processor designed for writing and editing large or complex documents, including structured documents.
Owned features - Online Review, Product Update Notification System, Logging functionality, etc.
Formulated Adobe Document Cloud based Online Review workflows in FM.
Took initiative to develop Comments Review Panel in React.js, the first of its kind in FM.
Refined comments import algorithm to facilitate import of comments from PDF to XML document.
Designed an algorithm to dynamically filter content in HTML5 nested lists.

React.jsGitHubC++Object-Oriented Programming (OOP)Algorithm DesignAlgorithm Development

Product Development Intern

May 2017 – Jul 2017 · 2 mos · Delhi, India · On-site

Created Selenium based monitors for real time monitoring of Adobe Target.
Automated deployment tasks for monitors via NewRelic Synthetics API on python.
Explored Capacity Planning. Automated data extraction process from multiple systems. Explored linear regression, ARMA model to fit data on Azure ML platform.
Technologies used : WebDriverJS, Python, NewRelic API, Microsoft Azure.

Hanyang university

Research Intern

May 2016 – Jul 2016 · 2 mos · Seongdong-gu, Seoul, Korea

Type-2 Fuzzy Logic applications in Machine Learning .
Worked on Fuzzy Clustering algorithms, Type-1 and Type-2 Fuzzy membership functions.
Modified Gustafson Kessel (GK) algorithm with Interval Type-2 Fuzzy approach and got improved results for classification problems and Image segmentation.
Technology used : MATLAB.

Python (Programming Language)