Yu Xiang
✰ Past: Infra / Cloud Eng / Android @ByteDance ;
✰ Present: Generative AI Enthusiast / K-line Enthusiast / @qodo.ai.
✰ Past: Infra / Cloud Eng / Android @ByteDance ;
✰ Present: Generative AI Enthusiast / K-line Enthusiast / @qodo.ai.
Contributed to CodiumAI, an Israeli AI startup ($11M Seed in 2023 & $40M Series A in 2024), as an open-source developer. Participating in the design, development, maintenance, and deployment of two core projects: PR-Agent and Cover-Agent. These tools leverage AI-driven solutions to enhance the efficiency and automation of continuous code review and continuous testing.
Key metric: for every 1K lines of code (additions/edits), Qodo Merge had on average 6.67 suggestions implemented.
Responsible for engineering productivity: optimizing development workflows and improving release quality by establishing engineering standards, managing releases, and shifting quality checks earlier in the development lifecycle (shifting left).
Authored and enforced a weekly release SOP, defined clear entry/exit gates from requirement to full rollout, unified A/B, hot-fix, and skip-release policies, resulting in ≤ 3 skipped releases and ≤ 30% delays in 24Q1, maintained a minor-release ratio of ≤ 10%, and cut code-freeze from 7.67h ‣ 5h.
Built an in-house productivity suite: batch MR, auto-merge to accelerate CI/CD, call-stack-based auto-attribution of issues in APM, and IM bots for chat creation, alerts, and trunk stability dashboards. This suite boosted for BM/On-Call efficiency by 61% (14 PD ‣ 6 PD).
Responsible for 3 product lines within the Quality Infra, including annual planning, product design, technical implementation, and facilitating cross-team collaboration and KA business integration.
Performance Analysis System:
Client-side (ByTrace): Developed to prevent performance regressions during the development lifecycle, it features automated detection of metric degradation, provides two profiling tools for single-trace (Tracery) and aggregate comparison analysis (Comparison View), and incorporates an automated service for root cause attribution.
Server-side (ByteHunter): An AIOps solution providing alarm suppression and root cause analysis.
A Typical User Journey (Trace Comparison): "Receive degradation detection ➡️ View root cause analysis conclusion ➡️ Drill down for detailed analysis ➡️ Confirm the responsible code submission"
Testing Services (App Competitor Evaluation): Delivered platformized, automated evaluation services, featuring two self-developed, general-purpose UX benchmarks: one for time-based performance analysis using high-speed cameras and video frame analysis, and another for resource-based analysis collecting perf metrics (e.g., CPU, Memory) via PerfAgent.
Intelligent Testing (SmartEye): Owened code coverage infra and leveraged it to implement precision testing.
Product:
UC Browser Mini
InfoFlow feature development: participated in business architecture refactoring and led the design and implementation of the Router component.
Optimized app performance, KPI improvements in crash rates, UI responsiveness, and launch times.
Independently delivered two innovative short-video applications targeted specifically at the Indonesian market.
Flyme Performance Team Member
Identified and drove AOSP-level performance optimization initiatives.
Developed and delivered new system-level features for Flyme OS.
Maintained the gfx stack, managed SurfaceFlinger and HWUI codebases, and resolved graphics-related issues on MediaTek and Samsung platforms.
Served as performance owner for selected models, kept a comprehensive optimization checklist, and ensured every item was validated in production.
Communication and Information Systems, GPA: 87.4
Information Engineering, GPA: 88.6
Delivered a technical talk Alibaba's "匠心堂" on Android Graphics.
Mobile App Testing at Scale - Regression and two performance analysis tools
Android performance Overview at ByteDance Infra, covering optimizations in Douyin and Toutiao, and OS-level improvements.
Color Mix Buddy
A simple and user-friendly online color mixing tool that supports various mixing algorithms and color selection options. Features multiple color mixing algorithms, built-in palette with predefined colors, and mix history feature.
Fancy Text Formatter
This is a modern web-based tool that transforms English text into stylish Unicode fonts—such as Bold Serif, Bold Sans, Italic Serif, and Italic Sans—within a clean, Apple-inspired user interface. It preserves all non-English characters, including Chinese, punctuation, spaces, and line breaks, ensuring seamless multilingual input.
Perfetto Plus
I extended Google's Perfetto project to support the Nanoscope protocol—an extremely low-overhead method-tracing solution for Android. By modifying the ART, Nanoscope traces Java method calls with minimal performance impact, incurring only around 10 milliseconds of overhead even when tracing millions of method calls on the main thread.
The tracing strategy is as follows:
The output format is a simple, human-readable push/pop structure, pattern:
[timestamp]:[methodSignature | "POP"]
Why not Google Trace Event Format because it proved excessively verbose and inefficient at handling millions of trace events, slowing both the generation and parsing of trace files.
Video Transcribe
This project uses OpenAI Whisper to generate high-quality subtitles from local video files or YouTube links, supporting Chinese and other languages, with built-in 360p video download, WAV audio extraction, optional timestamps, and a simple, flexible CLI.
The model will be downloaded automatically on first run and cached in the default directory ~/.cache/whisper.
Chrome Extension: Enhanced Screenshot
A browser extension designed to enhance screenshot functionality with beautiful backgrounds and customization options.
Day Trading @ U.S. Stock Market
I have a Bear blog where I collect ideas on technology, product, and AI learning.
