# Deng Liu

> Building AI Infra @ Sesame | ex-{Celonis, Snap, Twitter, VMWare, IBM}

Location: Mountain View, California, United States
Profile: https://flows.cv/deng

- 20 years of experience in building scalable data infrastructure, distributed systems, and storage platforms across top tech companies.
- Expertise in cloud-native architecture, Kubernetes, CI/CD, observability, and multi-cloud (AWS, Azure).
- Strong track record of leading high-impact projects, improving reliability, developer velocity, and cost efficiency.
- Holder of 10+ patents in storage, virtualization, and distributed systems.

## Work Experience
### Software Engineer @ Sesame
Jan 2025 – Present | San Francisco Bay Area

### Senior Staff Software Engineer / Engineering Manager @ Celonis
Jan 2022 – Jan 2025 | Palo Alto, California, United States
Tech Lead and Manager of the Data Infra team in delivering reliable and cost-efficient infrastructure to support the next-generation process mining query engine. I have initiated and driven several infrastructure projects across key areas, including:
- Elastic Process Mining Data Infra
- CI/CD & Release Management
- End-to-End Testing Infra
- Observability
- Backup and Restore
- Security and Compliance
- Cost Efficiency 
- Configuration Automation
- Production Issue Handling
- Miscellaneous Tools

### Software Engineer @ Snap Inc.
Jan 2017 – Jan 2022 | San Francisco Bay Area
Tech lead on several fun projects:
- LEO Tooling 
- Snap Minis Payment 
- Story Kit
- Sticker Kit
- API Platform
- Story API
- Snap Search
- Thumbnail Generation

### Software Engineer Technical Lead @ Twitter
Jan 2013 – Jan 2017 | San Francisco Bay Area
- Lead the effort of designing and implementing Twitter blob store distributed system. This system is used for storing photo, videos in Twitter.

- Worked closely with devops engineers to manager a distributed cluster of 80K nodes and delivery a reliable storage service for other Twitter product team

- Designed the shard management and topology management system as a generic self-contained system for Twitter’s distributed system.

### Senior Member of Technical Staff @ VMWare
Jan 2011 – Jan 2013
work on VMware ESX server kernel development:storage virtualization, filesystem, LVM, buffer cache, flash virtualization

### Storage System Research Staff @ IBM
Jan 2010 – Jan 2010 | San Jose, Ca
- Work on the data backup research project of IBM NAS Storage. This project improved the performance of filesystem(IBM GPFS) level data backup on IBM NAS Storage product in orders of magnitude

Tools: Linux kernel module, Kprobe, Linux audit sub system, C/C++, profilling

### Research Assistant @ University of Wisconsin-Madison
Jan 2009 – Jan 2010
I/O Path Optimization inside Linux Kernel

- Use profiling tool to pinpoint the hot spots along with the I/O path inside Linux Kernel 

- Design and implement a Dynamic Pagechunk Detecting Mechanism to improve the I/O performance

- With the I/O device, such as Flash, RAID, Fiber Channel, getting faster and faster, the overhead caused by transferring data between kernel and user can not be ignored as before. The old design of Linux kernel doesn't take effort in optimizing the kernel-user data copying, and it uses page(4kB) granularity to transfer data from kernel to user in a read() request. 

- To reduce the overhead of copy_to_user, I implemented a mechanism, which I call it pagechunk, in Linux kernel, which uses 64kB-unit to copy data between kernel and user. Pagechunk mechanism intercepts the read request, checks if the request is 64k bytes long and handles it by my code with 64k bytes copy_to_user.  By coalescing small copies into big copy, there was 15.41% improvement for "seq read w/ data in mem", and 20.17% and 30.84% improvement for "rand read w/ data in mem" and "rand read w/o data in mem" respectively. For "seq read w/o data in mem", there was a degradation in the performance, 2.19% decrease in the read rate.

Tools: Linux kernel, C,  GCC, Quilt, Make, Oprofile, Systemtap

### System Software Engineer @ IBM
Jan 2007 – Jan 2009
IBM Enterprise Storage Server (ESS) development team

- Developed and maintained ESS Microcode-OS sub-components: IOCTL, Kernel Pipe, File I/O

- Developed and supported a variety of internally used programs and tools such as dump analysis, statesave tool (used for collecting and analyzing diagnostic data from the failure), etc.

- Developed and maintained RAS components (a kind of storage system services which is dedicated to the Reliability, Availability and Serviceability of IBM Storage product)

- Problem Analysis (PA) on LPAR side: uses a method -- "problem domain" -- to manage and diagnose the problems of storage systems according to the "fault location" and "root-cause"

- Tested the cache usage and performance on the platform

- Regression test
Tools: AIX, UNIX, C/C++, Java, Shell, Perl

### Software Engineering Intern @ Intel
Jan 2006 – Jan 2006
Intel Software Solution Group

-  Member of Intel Platform Administration Technology (iPAT) team

-  Designed and implemented an intelligent synchronization method when deploying a huge number of data to hundreds of host machines
-  Designed and implemented a flow-control mechanism applied in multicasting using UDP datagram

Tools: Windows, C++, C#, Microsoft SQL2000

### Software Engineering Intern @ Motorola
Jan 2005 – Jan 2005
Motorola Global Software Group

-  Member of the iDEN Virtual Radio Interface Service (VRIS) team (iDEN is based on a dual-core chip system, VRIS is the middleware of iDEN phone software architecture)

-  mplemented parts of the Session Layer of SmartIPC (a communication stack between processors and/or between tasks in the same processor)

-  Measured Transfer Time VS. Message Payload of SmartIPC and UDP Socket
Tools: Linux, C, Clearcase


## Contact & Social
- LinkedIn: https://linkedin.com/in/deng-liu-b2379815

---
Source: https://flows.cv/deng
JSON Resume: https://flows.cv/deng/resume.json
Last updated: 2026-04-10