
Tao Xin
Senior Staff Software Engineer at Google
Highlights
- Seasoned programming enthusiast with proven track record in big data infrastructure, cloud storage, full-stack serving. Passionate learner of neural networks and LLM.
- Author of world’s smallest reactive UI framework - VanJS (https://vanjs.org/), a JSX-free, React-like web framework with 50x smaller bundle size and much better performance.
- 16-time Perfy Award winner (a Google-wide award that recognizes notable performance optimization projects), serving Perfy Award committee for 5 consecutive terms.
- Top-notch competitive programmer: 8th place in 2007 ACM-ICPC World Final.
- Programming languages (ordered by fluency):
C++, JavaScript, Python, Lua, Java, Rust, Kotlin, C#, bash, elisp, awk, etc.
Professional Experience
Google Play Longitudinal Data Infrastructure (1/2023 ~ Present)
- Role: Chief Architect / TLM of Civita, a general purpose data infrastructure that transforms year-long user activities into business insight.
- Business impact: Elevating Google Play’s business focus from app acquisition to post-install activities and life-cycle management, where most of user spending occurs.
- Scale: Hundred petabytes of data across variable data sources. Tens of processing nodes.
Google Cloud Storage: Data Plane (7/2021 ~ 1/2023)
- Role: Senior Engineering Manager. Grew the team by 2X to 20+ HC. Developed next generation leaders.
- Responsible Areas: data serving, data caching, data replication, SLO, upload planning, integration w/ Colossus file system, etc.
- Notable Product Launches: Anywhere Cache, Turbo Replication.
- Infrastructure Overhaul: shard service modularization, CFS2 migration, spindle-weighted upload, speculative hedging, 2.1 data striping.
- Operational Excellence:
- Refactored oncall / onduty process for the 200+-people GCS serving org.
- Cut down onduty bug backlog by 3x.
- Reduced onduty workload from 80%~100% to ~50%.
- Cut down p90 presubmit running time by 4x (28m -> 7m), with <1% flakiness.
- 100% favorable manager survey.
Big Data Platform for Google Play & Android (12/2015 ~ 7/2021)
- Role: Chief Architect / TLM of Chipper, the unified logs processing, metrics, attribution, experimentation and ML data platform for the entire Google Play & part of Android.
- Business impact: Provide vital real-time datasets for 3.25 billion monthly active users with a $42 billion annual run rate.
- Scale: Over 5+ years, I significantly scaled out the infrastructure:
- Input: 36M events / sec from 100+ data sources. ~300+TB / day.
- Daily output: ~500TB in 200+ tables, 1,700+ metrics, 1,100+ dimensions, 400+ custom fields.
- Median data latency: 5 minutes (low-latency tier), 50 minutes (regular tier).
- Managing a large codebase with 700+ unique contributors.
- Powering 1,200+ FTEs across 8 offices in 5 timezones.
Dynamic Display Ads, Google (10/2011 ~ 12/2015)
Notable achievements:
- GPA Build Unification: Built the next-gen unified offline data processing and index building system for Dynamic Display Ads ($550M annual run rate in 2015), replacing 5 legacy codebases from various Ads products with a single, consolidated, vertical-independent, config-driven and extensible platform.
- Dynamic Ads Layout Optimization: Smart layout engine that enables ads serving system to choose the best performing layout on an impression-by-impression basis. Patent GP-17175-00-US: Automatically explore, infer and choose the best performing layout on an impression-by-impression basis.
- GPA Big Data: Scaled out the index building system to serve 200+ million feed items uploaded by numerous advertisers: 10X improvement over prior generation. This enabled the onboarding of many large customers and improved conversion rate by 4%.
Commerce Transaction Platform, Microsoft (7/2009 ~ 9/2011)
Notable achievements:
- High Potential Employee for FY2012 (Top 4% performers).
- DQL - Distributed Query Language: Designed and implemented a new programming language - DQL, which is based on SQL, with a rich set of feature extensions that enabled cross-server, cross-partition, cross-component data query, extraction and transfer.
- Mercury - Asynchronous Event Processing System: Built the core platform for asynchronous message communication. Also built asynchronous database writer, general notification engine and offline risk evaluator on top of it. This is one of the essential projects to launch Windows Store.
- Transformation Engine: Implemented a powerful domain-specific language interpreter for decision scientists to configure sophisticated models for risk management purposes.
Personal Open Source Projects
VanJS
- Smallest reactive UI framework in the world, offering comparable developer experience as React with 50x smaller bundle size.
- 3.7k GitHub stars.
- 14th-most starred GitHub repo among all front-end frameworks in 2023.
twikoo-cloudflare
github.com/twikoojs/twikoo-cloudflare
- The cloudflare deployment for twikoo comment system, with much shorter cold start latency (6s -> 0.5s) compared to other deployments.
Tech Talks
I hosted numerous tech talks to a wide variety of audiences across Google. Notable topics are:
- An introduction to Chipper
- Deep-dive: scalable real-time logs processing
- Deep-dive: logs processing optimizations
- What we learned from Streaming Flow (foundation infrastructure for real-time logs processing)
- C++ optimization tips
- Protocol-buffer tips and tricks
- Pitfall of 2-level hash tables in C++
Education
B.E. in Computer Science, Shanghai Jiao Tong University, 2005~2009:
- 8th place in 2007 ACM-ICPC World Final (among 5,606 teams around the world).
- 2007 National Scholarship (Top 2% students).
Trivia
- I learned LLM by reading code from https://github.com/karpathy.
- I enjoy supercharging my productivity and concentration by building extensions to work environment, including:
- Chrome extensions
- VS Code
- Hammerspoon scripts
- Karabiner Elements config
- Emacs scripts (obsolete)