From First PR to Maintainer: Field Guides from Nine Taiwanese Open-Source Contributors

Most software engineers rely on open-source tools every day, but the jump from user to core contributor is often steep: monolithic codebases, intricate architectures, and the friction of asynchronous, cross-timezone collaboration.

The inaugural 2025 IT Matters Awards Open Source Community Contribution Award brought together nine prominent Taiwanese open-source practitioners. Spanning Kubernetes, distributed AI computing, data pipelines, the Linux kernel, open government, and foundational input methods, their work illustrates four distinct postures in open source:

Deconstructing their practices and mental models reveals how engineers can find viable entry points, build the trust required to step up as project maintainers, and sustain the endurance needed for a journey spanning years—or even two decades.


From Perimeter to Core: Three Entry Points into Massive Upstream Projects

When engineers confront massive global projects, jumping straight into the core scheduler or underlying engine often ends in frustration.

“Upstream” refers to the official source repository, as opposed to downstream branches forked locally by individuals or enterprises. Only by contributing directly upstream do patches and designs land in the main trunk for global adoption. Wei Lee, Han-Ju Chen, and ChengHao Yang each demonstrate a distinct breakthrough strategy.

From Docs Localization to Core Architecture: Wei Lee’s Progressive Path

Wei Lee, a software engineer at Astronomer, is a Project Management Committee (PMC) member and Committer for Apache Airflow. In the Apache Software Foundation’s (ASF) rigorous governance model, Committers hold direct write access to the codebase, while PMC members oversee personnel promotions and hold decision-making votes on major project direction.

His Airflow journey started as the Code Owner for Traditional Chinese documentation and localization. Serving as the gatekeeper for that directory allowed him to master the project’s sprawling repository structure, CI test conventions, and asynchronous global communication cadence.

Using this role as a launchpad, he began tackling sample code errors in the docs and addressing small code issues. Once his footing was secure and community trust had accumulated, he ventured into core architectural work.

AIP-90 shipped in Airflow 3.1. Wei served as the lead for the improvement proposal AIP-90: Human-in-the-Loop and drove the design and implementation of sub-proposals AIP-74 and AIP-76 under the Expanded Data Awareness umbrella—enabling schedulers to track dataset states and structural mutations—as well as migration tooling. On the back of consistent, reliable engineering output, he was elected as a Committer in 2024 and subsequently elevated to the PMC.

High-Intensity Routine and Sheer Engineering: Han-Ju Chen’s Technical Breakthrough

Han-Ju Chen took the direct engineering route. When he started contributing to Flyte, a cloud-native workflow orchestration engine, as a college student, he had virtually no production experience with Go or hands-on Kubernetes administration.

To clear the learning curve, he systematically ran official examples, studied architectural documentation, and tackled good-first-issue tickets. To counter the trial-and-error uncertainty and burnout risk common in early open-source work, he enforced a disciplined cadence of six hours a day, six days a week, punctuated by scheduled exercise and rest. After shipping over 50 PRs within 21 weeks, he was elected as a Flyte Committer in November 2023.

The public track record of a Committer on a major open-source project serves as a hard-to-fake technical endorsement. His deep hands-on expertise in Kubernetes controllers and cloud-native scheduling architectures carried directly into his day job.

He was subsequently recruited by commercial open-source company Anyscale as a software engineer and went deep on KubeRay—the flagship project for running Ray distributed computing natively on Kubernetes. There, he led multi-cluster support, AI inference service integration, and large-scale scheduling optimizations, culminating in a featured presentation at KubeCon Europe.

Bypassing the Code Wall: ChengHao Yang’s Release Team Strategy

For engineers looking to enter the Kubernetes ecosystem without getting lost in its monolithic codebase, ChengHao Yang demonstrated a strategic non-code entry point: the Kubernetes Release Team Shadow Program.

Kubernetes ships three releases a year (roughly every four months). Each cycle is driven by a Release Team of around 30 global volunteers, organized into sub-teams covering test signal monitoring, bug triage, docs, and release notes. Each sub-team is led by an experienced lead mentoring a handful of openly recruited “shadows” (official guidance caps shadows at around four per role).

Yang joined the Release Signal team starting with Kubernetes v1.32, working his way up from shadow to sub-team lead. This pathway yields three distinct advantages for newcomers:

Monitoring release signals gave Yang a bird’s-eye view of Kubernetes’ overall architecture and common failure points. Equipped with cross-team visibility and CI fluency, he knew exactly where to claim meaningful architectural work.

Yang not only became a core maintainer for the deployment tool Kubespray, but also pushed code directly into the upstream kubernetes/kubernetes repository—for example, claiming a piece of the architectural overhaul that the relevant SIG is driving across the board: refactoring the Service CIDR Controller to introduce contextual logging, substantially improving control plane debuggability and observability.


From PR to Maintainer: Why Reviewing Code Builds More Trust Than Writing It

In large-scale open-source projects, every line of merged code commits the team to years of backward compatibility, test reliability, and technical debt. New code is often a long-term liability, whereas high-quality code review is an immediate asset that relieves maintenance strain. Maintainers inundated with hundreds of PRs daily are starved for one thing: trusted reviewers who can act as dependable filters.

Han-Ju Chen’s GitHub track record reflects this dynamic: as of November 2026, he has authored more than 470 PRs and completed more than 780 reviews—roughly 1.6 times his PR count. Similarly, ChengHao Yang took on an extensive review load in Kubespray before earning official reviewer and maintainer status.

Authority as Retroactive Recognition: The Five-Rung Kubernetes Contributor Ladder

In open-source organizations that prize transparent governance, this progression from submitting code to shouldering maintenance duties is codified into formal governance. Take the Kubernetes community’s official Contributor Ladder, which structures permissions and responsibilities into five clear rungs:

RoleThresholds & QualificationsCore Privileges & Responsibilities
ContributorSign the Contributor License Agreement (CLA)Start via good-first-issue or help-wanted tags to fix docs, add tests, or file issues
Organization MemberMultiple substantive contributions; sponsored by 2 reviewers or approvers from different companiesGitHub organization membership; permission to assign issues and trigger automated CI runs
ReviewerDeep domain expertise in a specific module; sustained history of thorough PR reviewsListed in module OWNERS as reviewer; checks implementation details, test coverage, and code correctness
ApproverActive Reviewer for at least 3 months; primary reviewer on 10 significant PRs; 30 PRs reviewed or mergedListed in module OWNERS as approver; gates backward compatibility, maintenance burden, and design direction
Subproject OwnerSustained technical leadership steering the subproject’s trajectoryFinal accountability for long-term roadmap, release cadence, and architecture; nominates new approvers

Under this two-tier review model, repository OWNERS files separate reviewers and approvers. A reviewer verifies the implementation details and posts /lgtm (“looks good to me”), while an approver validates architecture and compatibility before issuing /approve. The automation bot merges the PR only when both signals are present. This separation of concerns prevents senior architects from drowning in line-by-line diffs while delegating review responsibilities across a broader circle of trusted peers.

In contrast to the CNCF’s fine-grained, five-rung ladder, Wei Lee’s home base—the Apache Software Foundation (ASF)—runs on two-tier governance: Committers with direct write privileges, and PMC members who steer project direction and vote on personnel matters.

Though the terminology and tiers differ, the underlying principle is identical: authority is the retroactive formalization of existing behavior. Engineers consistently perform the responsibilities of a tier long before the community officially confers the title of Reviewer or Committer upon them.

Delivering High-Value Reviews Without an Official Title

In corporate engineering, code reviews are typically reserved for team leads or senior peers. In open source, however, every PR is an open public discussion. Even without official standing, any contributor who validates a PR against real-world test environments directly lifts the heaviest burden off maintainers:

TIP

Three common forms of substantive review contribution:

  • Verify boundary and edge conditions: Test how the proposed change behaves under malformed inputs, high concurrency, network partitions, or unexpected node restarts.
  • Audit test coverage: Check whether changes miss crucial unit tests or end-to-end (e2e) integration suites.
  • Run local reproductions and paste logs: Pull down the PR branch, compile it locally, run test suites, and share the exact environment and output (e.g., confirming that a bug is fixed on a local KinD cluster, while flagging that it triggers a panic under IPv6 setups, accompanied by full stack traces).

When a contributor consistently delivers reproducible local validation and edge-case testing, maintainers come to recognize the contributor’s judgment through day-to-day collaboration—and take the initiative to nominate them for the official reviewer roster.


When Existing Tools Fall Short: From Filling Gaps to Twenty-Year Digital Infrastructure

Beyond joining established global giants, another cohort of engineers chooses to build their own tools when confronted with unmet needs in their toolchains or civic environments. This journey tests two distinct capabilities: pinpointing genuine friction points upfront, and sustaining maintenance across decades.

Zero-to-One Breakthroughs: Solving Kernel Performance and Cultural Context Gaps

Focusing on eBPF, the Linux kernel, and container workload orchestration, Ian Chen observed that when Kubernetes clusters face CPU saturation or extreme workloads, latency-critical applications—such as the free5GC 5G core network implementation—can suffer sudden performance degradation.

Leveraging the sched_ext framework introduced in Linux Kernel v6.12 (which enables custom user-space CPU scheduling policies via eBPF) alongside Go, Chen built Gthulhu from the ground up. This cloud-native runtime scheduling platform translates Kubernetes workload intents into low-level Linux task priorities to shield critical paths, with the long-term goal of entering the CNCF Sandbox.

In the generative AI space, Liang-Hsun Huang, founder of Twinkle AI, spotted a linguistic and cultural disconnect. Mainstream open-source large language and multimodal models (such as the Llama family) lack Taiwan-localized corpora during training, and routinely misjudge Taiwan’s regulations, everyday knowledge, and cultural customs—such as mistaking traditional ceremonial flower wreaths for traffic warning signs.

Huang launched the Twinkle AI community to bridge this gap. Building on the Meta Llama 3.2 3B architecture combined with Taiwanese Chain-of-Thought (TCoT) reasoning data, he trained and open-sourced Formosa-1-an on-device Traditional Chinese reasoning model whose base weights carry an MIT license.

His team has also compiled datasets including FineWeb-edu-zhtw and Formosa Vision, alongside the Twinkle Eval evaluation framework—rounding out the open base for Traditional Chinese AI across data, model weights, and evaluation standards.

Sustaining Daily Utilities for Decades: zonble and Ronny Wang

When home-grown projects target foundational daily utilities or public civic goods, the engineering challenge shifts from initial architectural design to intergenerational maintenance stamina.

Back in 2004, when macOS still offered a deeply frustrating native Traditional Chinese typing experience, Wei-Chung Yang (zonble) went to work on the architectural design and module development of OpenVanilla, which supports a wide range of input needs. He later contributed to the core engine of Yahoo! KeyKey.

Over the course of two decades, he has weathered endless OS overhauls and breaking API migrations, authoring and maintaining dozens of polished Cocoa, Objective-C, and Swift libraries across his blog and GitHub—foundational craftsmanship relied upon by generations of developers.

Ronny Wang, founder of OpenFun Ltd. and a core g0v contributor, set out in late 2012 on an open-data marathon that has now lasted more than a decade: frustrated by anti-scraping measures on the Ministry of the Interior’s property transaction portal, he wrote a scraper himself and published the data.

He has gone on to build and maintain pivotal platforms including News Helper, Taiwan Company Registry, Open Campaign Finance, and Open Tender. To keep these non-profit civic tech initiatives running on minimal budgets, he even engineered hisoku, a custom micro-PaaS platform.

Wang once likened untangling locked-down government PDFs and chaotic web pages to “battling monsters in a video game”—all with the goal of turning records that are public in form yet practically unsearchable into APIs and structured databases anyone can query.


Beyond the Individual: Local Community Networks That Lower the Barrier to Entry

Whether attempting to contribute to international upstreams or maintaining local projects, beginners are most frequently paralyzed by isolation—not knowing where to begin. Examining the trajectories of these nine awardees reveals a common backbone: an interconnected, multi-generational local community support network.

In the summer of 2012, Denny Huang, then a student, was profoundly inspired after attending COSCUP (Conference for Open Source Coders, Users, and Promoters). He wondered: “If adults can host conferences to share their work, why can’t we build a conference where students can take the stage to share their own projects and passion?”

In 2013, he and his peers co-founded SITCON (Students’ Information Technology Conference), with Huang serving as lead organizer for its first two editions. SITCON opened a door for teenagers and undergraduates to bypass exam-driven educational tracks and immerse themselves early in Linux, Git, and the ethos of open-source sharing.

As students transition into industry or specialize in distinct technical domains, the baton passes to established professional communities: PyCon TW, COSCUP, and the Cloud Native Taiwan User Group (CNTUG).

As an official CNCF Ambassador, Phil Huang has long steered CNTUG, bringing KCD Taiwan 2023 and KCD Taipei 2024 to fruition locally. In 2023, he initiated the CNCF Glossary Traditional Chinese Localization Project, lowering the terminology barrier for Chinese-speaking developers entering the cloud native ecosystem.

This ecosystem fosters a self-reinforcing flywheel: ChengHao Yang, once an attendee in the audience, later stepped up as a CNTUG co-organizer and Lead Organizer of KCD Taipei; Wei Lee mentors newcomers in communities like “Open Source for You” (源來適你), helping them navigate setup hurdles and pull request workflows. Through this lineage, more Taiwanese engineers follow their predecessors’ paths into international core circles.


How Long Can Passion Last? Professional Leverage and Mental Models for the Long Run

In large-scale projects, a PR drafted over three late nights might be tagged with needs-rebase by an automation bot and left untouched for two weeks; a senior reviewer might leave a dozen exacting change requests requiring a complete refactor; or a hard-won feature might be scrapped wholesale during an architectural review meeting the following week due to shifting project priorities.

When day-to-day collaboration is this high-friction, relying solely on spare-time passion or the short-term motive of padding a résumé with an open-source title rarely sustains high-output contributions past the three-year mark. The ability of these nine practitioners to remain engaged for years-or decades-rests on two foundational pillars: professional leverage that makes full-time work and open source mutually reinforcing, and a mindset that views open source as compound interest on deep technical understanding.

Four Archetypes of Professional Leverage

  1. Transforming production pain points into upstream PRs: When SREs or backend engineers encounter upgrade failures, memory leaks, or missing logs in production, maintaining a bespoke internal fork creates compounding debt. Instead, as demonstrated by ChengHao Yang, generalizing the fix and upstreaming it eliminates private fork maintenance while solving the company’s problem and establishing an international engineering track record.
  2. Joining Commercial Open Source Software (COSS) companies: For Han-Ju Chen at Anyscale (the company behind Ray) and Wei Lee at Astronomer (the commercial steward of Apache Airflow), shipping upstream features and conducting code reviews are explicit job expectations and performance metrics. Professional career trajectories and open-source contributions become entirely aligned.
  3. Aligning enterprise architecture with developer advocacy: For Phil Huang, a Senior Cloud Solution Architect at Microsoft Taiwan with prior tenure at Red Hat and VMware, guiding enterprises through Linux and Kubernetes adoption is core business. Enterprise field lessons feed directly back into the CNCF community, while strong community visibility continuously reinforces trust with enterprise clients.
  4. Sustaining civic projects through commercial business models: To prevent civic data platforms from burning through personal goodwill, Ronny Wang founded OpenFun Ltd. in 2023 to explore commercial data service models—“making open data a legitimate enterprise”—fueling the sustained maintenance of public digital infrastructure through steady revenue.

Open Source as Compound Interest on Technical Understanding

Beyond practical leverage, three shared engineering mindsets sustain their multi-year commitment: