Case studies
Governance, infrastructure, and product leadership — examined the way I'd want a research method examined: what the problem actually was, what we tried, what it cost, and what I'd tell someone doing it next.
Lezismore: Building Rules Beyond Platforms
How do you govern a digital space without becoming the very thing you wanted to escape?
The Challenge
Mainstream social platforms have never been safe spaces for Taiwan's LGBTQ+ community, especially for sex-positive discussion, queer storytelling, and marginalized identities. Censorship, algorithmic suppression, and the constant threat of takedowns made it clear: we couldn't rely on existing platforms to host our narratives, depend on external moderation to protect us, or wait for "better policies" to happen. If we wanted a space that worked for us, we had to build it ourselves.
The Solution
I built Lezismore: an independent, privacy-first LGBTQ+ community where users could discuss sexuality, identity, and experience without risk of deplatforming. Privacy-first design (no surveillance, no forced real-name policy), decentralized governance (no single authority dictating rules), and content moderation without surveillance — balancing safety against autonomy in a space often targeted by bad actors.
Trade-offs
What worked: anonymity and privacy let users express themselves without fear of exposure; a real, self-sustaining community emerged; the platform escaped the sudden takedowns and corporate bias of mainstream moderation.
What didn't: no real-name policy meant handling bad actors was harder than simply banning accounts; running a non-monetized independent platform carries a high operational burden; and with no central authority, we had to navigate real power struggles over who makes the rules.
Lessons
Autonomy comes with a cost — you can't demand freedom without also taking responsibility for governing it. Moderation without surveillance is possible, but requires a cultural shift, not just better policy. Building an alternative is harder than criticizing the status quo: creating a space is one thing, sustaining it for a decade is another.
Building Without Permission
How do you build resilient infrastructure when you can't rely on traditional gatekeepers?
The Challenge
When you rely on corporate platforms, you don't own your community — you rent it. Rules change overnight, moderation policies shift, and entire spaces disappear without warning. True autonomy meant building from the ground up, without asking permission.
The Solution
An independent, self-hosted ecosystem built on open-source tools: hosting migrated from Digital Ocean to GCP to AWS; core services on Docker, Nginx, WordPress, Discourse, n8n, CodiMD, and Bookstack; custom backup and recovery pipelines; automation via webhooks and dashboard integrations. The goal wasn't just to self-host — it was to build infrastructure that could survive long-term.
Trade-offs
What worked: full control over platform rules with no external moderation dictating policy; scalability through modular, open-source components without vendor lock-in; automation that reduced repetitive maintenance.
What didn't: no customer support — when something broke, we fixed it ourselves; an ongoing maintenance burden of updates, patches, and monitoring; constant interoperability friction between open-source tools never designed to work together.
Lessons
Owning infrastructure means owning its failures. Redundancy isn't optional. And automation isn't about convenience — it's about survival. Good infrastructure, like good governance, is invisible when it works and painfully obvious when it fails.
Autonomy Is Built, Not Given
How do you foster self-governance without descending into chaos or majority-rule oppression?
The Challenge
Most platforms offer two extremes: authoritarian moderation or mob rule. Neither works for a marginalized community trying to build real autonomy — mass reporting enables brigading and silences unpopular voices; no oversight lets harassment and bad actors thrive; rigid enforcement fails to adapt to context.
The Solution
A structured, community-driven moderation model designed to prevent abuse while staying fair: context-based reporting that required specifics on what rule was broken and what harm was caused; weighted, pattern-based review that flagged coordinated mass-reporting attempts rather than treating report volume itself as evidence; anonymized governance audit logs for accountability and policy refinement; and cooling-off periods before major actions, to prevent knee-jerk decisions.
Trade-offs
What worked: brigading and weaponized reporting stopped working; users engaged with governance instead of defaulting to outrage; trust in the process increased because enforcement was visibly structured rather than arbitrary.
What didn't: genuine community-led governance takes real time and engagement from users; some detection methods (e.g. IP-pattern analysis) couldn't be disclosed without teaching people to evade them; and some users still expected the instant justice of a ban-first system.
Lessons
Mass reporting isn't justice — treating reports as data points rather than evidence prevents majoritarian abuse. Transparency has to be strategic: full disclosure of enforcement mechanisms can weaken them. Community autonomy doesn't mean no rules — it means shared responsibility.
Editorial, Platform Design, and Digital Advocacy
How do you build a storytelling ecosystem that resists shallow visibility and prioritizes depth, impact, and autonomy?
Storytelling isn't just content — it's the conditions that let stories exist, grow, and shape the world. For nearly a decade, my approach has had three parts: shaping narratives that counter erasure and misrepresentation; building infrastructure that lets diverse voices emerge and multiply; and deciding, deliberately, who tells stories, where, and how — choosing strategic visibility over algorithmic virality.
Platform design beyond surface engagement
A split between main site and community hub let users engage at different levels to protect anonymity while fostering real interaction. Deliberate friction, not the instant-swipe logic of dating apps, favored depth over speed. UX and community language were adapted — not just translated — to reflect the lived experience of Taiwanese queer users. And positioning stayed deliberately off the major platforms: no reliance on Instagram, Facebook, Twitter, or Threads, no influencer marketing, no mainstream visibility hacks. Most platforms optimize for reach; this one optimized for meaning.
Editorial work
An interview series ("Unseen Pride") profiled LGBTQ+ lives that don't fit mainstream molds — a Taiwanese drag king relocating to the Netherlands, an award-winning Taiwanese transgender writer building a life in Japan. A recurring Community Observation series turned attention inward, onto the internal dynamics and struggles within queer community itself, rather than only external oppression. A self-produced podcast passed 30,000 listens; the platform's own writers produced 30,000+ community-authored articles and 90M+ views.
Trade-offs
Refusing the algorithm's game means slower growth and real, ongoing effort to sustain engagement without platform-driven virality — a conscious trade against the reach that playing by mainstream rules would have bought.
Leading a 28-Person Team & AI Product Governance
How do you balance leadership, compliance, and innovation while building AI-driven products in a cross-functional, high-stakes environment?
The Challenge
AI product teams live in permanent tension: innovation speed against legal constraint (GDPR, CCPA, PDPA); cross-team collaboration against siloed expertise across engineering, legal, and product; and the pull of data-hungry AI training against user privacy. I led a 28-person cross-functional team through exactly this tension.
The Solution
A compliance-driven AI product strategy that treated governance as a design constraint, not a limitation. One concrete case: an AI voice recognition feature raised the question of whether to store students' voice recordings to improve model accuracy. Rather than a binary choice, we transformed voice data into non-identifiable text tokens — preserving accuracy without storing sensitive audio, aligning with GDPR/CCPA/PDPA, and reducing storage costs.
Trade-offs
What worked: cross-team buy-in for governance-first AI design, built through clear risk assessment; proof that privacy-first AI can be both innovative and scalable; governance embedded in each team's decisions rather than imposed top-down.
What didn't: most competitors weren't holding themselves to the same standard, so compliance sometimes read as a competitive cost; some teams initially treated compliance as friction rather than strategy; and the privacy-first approach required real additional R&D investment to hold accuracy steady.
Lessons
AI governance is a leadership challenge, not only a legal one. Legal teams can't dictate AI ethics alone — it has to be built into product strategy. And regulation, treated seriously, can drive better design rather than only restrict it.