Every stop, closer to the real problem.
{{item.description}}
{{item.description}}
What shipped, and what it moved.
{{p.summary}}
{{p.outcome}}
{{p.summary}}
{{p.outcome}}
A no-code B2B2C SaaS platform (cand.xyz) where very different domains - community, commerce and blockchain, each with its own APIs - integrate into one system. Clients configure and run services for their own end-users across these domains, stitched together through canD's web, mobile app and admin surfaces.
At Canlab I owned canD end-to-end. I led a 10+ person cross-functional squad while simultaneously running domain-specific task forces across community, commerce and Web3, taking new opportunities from discovery through to go-to-market. I owned direct client relationships across 20+ clients, embedding structured user research as a standing practice alongside design.
The squad responsible for canD's core systems - web, app and admin - rather than any single domain: the widest-reaching work, spanning the whole product. I was trusted to lead it within 2 years of joining, owning the roadmap and running day-to-day delivery for a 10+ person cross-functional team, staying hands-on in execution while the platform scaled from 2 to 30+ client services.
Ran another task force alongside the core squad, this time focused on review-content-driven commerce for beauty and fashion clients - validating new revenue opportunities and shipping them into the live platform.
Ran a separate cross-functional task force alongside the core squad, exploring blockchain's role across identity and commerce - from non-transferable identity tokens to NFT-backed ownership and crowdfunding - validating new domain opportunities and taking them from discovery through integration and go-to-market.
Whether you're hiring or just want to talk product, I'd love to hear from you.
A node-based visual interface for building generative AI workflows (comfy.org), used by 4M+ people to wire together image, video and other AI models without writing code. Open-source at its core (Local/Desktop), with a paid Cloud offering for users who don't want to manage their own compute.
I joined ComfyUI at the founding of its support function as Cloud launched, building the support infrastructure from scratch. Beyond day-to-day support and QA, I built an AI workflow that automated resolutions across multiple channels, and worked hands-on with two of ComfyUI's LLM agent products - stress-testing Comfy Agent and contributing to the design of Docs Agent.
Designed the end-to-end support journey per user segment - deciding what could be caught upstream before it ever reached support, what could be self-serve versus direct manual support, spanning FAQ structure, article publishing, community-driven solutions, and direct support.
{{p.summary}}
{{p.outcome}}
Worked hands-on with ComfyUI's LLM agent products, deepening my familiarity with LLM-powered product design across both a builder-facing agent and a documentation agent. Details are confidential while these ship.
{{p.summary}}
{{p.outcome}}
Whether you're hiring or just want to talk product, I'd love to hear from you.
My MA in Service Design is grounded in participatory research, prototyping and iterative validation. Across three projects, I worked with people directly affected by the systems I was designing - trainees and operators at a coffee training social enterprise, venue owners and musicians navigating an entry experience, and people with lived experience of prison - to define the real problem before proposing a solution.
Ran participatory research with multiple stakeholders to uncover an untapped revenue opportunity, then designed and validated a new service through iterative prototyping.
Concept adopted; now in active development as a live service.
Service DesignIdentified friction in the entry experience for venue owners and musicians through 35+ interviews and concierge testing, then prototyped a simplified onboarding flow.
Won the Social Impact prize.
Service DesignDesigned how a paid programme for people with lived experience of prison could sustain trust, continuity and motivation within a constrained, bi-monthly format.
Service DesignWhether you're hiring or just want to talk product, I'd love to hear from you.
As canD's client base diversified, the existing block system couldn't support the visual variety clients needed. Clients were hitting configuration ceilings, limiting their ability to differentiate their services, and limiting canD's ability to win and retain them.
Led product development of Block System 2.0, a modular visual customisation system, alongside a Block Editor enabling clients to configure pages intuitively without engineering support. New visual-composition blocks (image, video, button, spacer, section, grid) replaced the old data-only blocks, and each component gained granular, atomised design settings.
Client acquisition doubled in 3 months, and onboarding time was cut by 54%. Business sales were unblocked as clients gained the flexibility to build advanced, differentiated designs, and the project laid the groundwork for a scalable, service-wide design system extending to navigation and footer areas.
As canD scaled to support clients running multiple sub-communities, app navigation became a growing source of friction. Client feedback and interviews revealed two critical issues: users couldn't orient themselves across communities, and DMs from sub-communities were being missed, creating churn risk for second-stage contracts.
Led discovery and product delivery with a team of 1 designer and 3 engineers. Ran interviews with four client companies across beginning, single-community and multi-community stages - asking them to simulate how they'd design their own app navigation if given customisation tools. The interviews confirmed the hypothesis and surfaced an unexpected second issue: missed DMs from sub-communities.
Shipped a navigation overhaul including customisable app navigation per client, a joined sub-communities page, and a unified DM inbox in the service hub.
Client acquisition doubled in 3 months, navigation-related support tickets dropped by 85%, and DM usage within sub-communities rose 120%. This project ran in parallel with the Block System 2.0 redesign, together removing the friction that had been blocking both new acquisition and second-stage sales.
In canD, Web3 community clients needed a way to represent identity, achievement, and membership that was permanent and non-transferable. Standard NFTs could be sold or transferred, undermining their credibility, and the UX was too complex for non-crypto-native community members.
Led a cross-functional task force to validate the opportunity, define the product, and integrate Web3 APIs into canD's live platform, shipping CryptoBadge - a non-transferable token system built on the Soul-Bound Token standard - with real client deployments across community, ticketing, and membership use cases.
This was one of canD's first Web3 integrations, validating a new domain opportunity and supporting client acquisition through tailored demos and MVPs.
Successfully shipped and validated across multiple real clients: Seoulbound (global Web3 event ticketing), Nonce (DAO community membership), Vake (social campaign platform), and Hypersportsclub (NFT holder recognition). Designed and demoed a live service at ETHDenver 2023.
canD's sign-up flow was built generically, with no way for clients to collect the specific information they actually needed from their own users at sign-up, no gate to approve or reject who joined, and no clear definition of what state a sign-up was in along the way. It began with a specific client's request, but surfaced a broader need and clear potential to expand sales as the product grew.
Expanded the sign-up flow to let clients collect additional, client-specific information from users at the point of sign-up, rather than only a fixed generic set of fields.
Added an approval/rejection step so clients could gate who actually joined their service, and defined clear states across the sign-up lifecycle - pending, approved and rejected - so both admins and users always knew where a sign-up stood.
Successfully acquired and launched the targeted client, and the feature was since adopted by 30%+ of existing clients too.
Beauty and fashion content communities had engaged audiences but no commerce layer: they could publish reviews and content, but had no way to turn that trust and attention directly into product sales for their own or partner brands, in a way that still felt native to a review-first, content-driven community.
Led discovery and build for a review-content-driven commerce platform on canD, shipping tailored storefronts for international beauty/fashion community partners.
Scope covered membership, category structure, product data and pages, review creation and curation, trial-group (체험단) features for seeding early reviews, and localisation - scoped differently per partner's existing user base and market.
Launched for targeted international partners, and drove 5+ additional client acquisitions within 3 months - validating review-content-driven commerce as a repeatable pattern on canD rather than a one-off build.
Creators making MOC (My Own Creation) products sold through licensors had to hand over licensing rights to get their designs manufactured and sold, losing direct control of their own revenue channel. Mass-produced copies also had no way to be reliably told apart from the original.
Buyers, in turn, had little trust in a creator's brand outside the platforms they'd issued through, and little visibility into whether a product would actually be delivered as promised.
Designed a blockchain crowdfunding model where creators could fund and sell a limited run of a physical product directly, with an NFT minted per unit as proof of ownership and pre-order - giving buyers a verifiable, transferable claim rather than just a licence.
Scoped and designed the full user and admin flow: crowdfunding page, wallet connection and PIN setup via a dedicated CAN Wallet, payment, NFT drop on successful funding, shipping request and tracking, and resale rights for a second-hand market after initial delivery.
Brickbit didn't reach commercial launch, but it proved the pattern: NFT-backed ownership and crowdfunding for creator-made products, buildable end to end. That groundwork contributed to a 150% increase in Web3 client acquisition within 3 months, the milestone the Blockchain TF was set up to hit.
I joined as a QA intern and, within about two years, ended up owning QA for the entire product team, testing every development ticket myself. As the team and product grew, this stopped scaling: the backlog of tickets waiting on me kept growing, and testing features I hadn't planned meant constant back-and-forth with the people who had, since I lacked the context they'd built up. Planning documents were rarely kept current either, since fast, verbal decision-making was how the team actually worked.
I went back to first principles: what is QA, and why do we do it? I redefined it as quality assurance, not testing, meaning the person best placed to own it is whoever understands the feature best, usually its own planner or PM, not a separate, centralised QA role.
I proposed shifting default testing ownership to each feature's own planner, with the QA role stepping in only when asked, and keeping ownership of bugs and maintenance work that weren't tied to a single feature. I presented this first to the planning team, then to the whole product team, and wrote it up as a shared philosophy document that defined what QA meant for the team going forward.
Post-QA bugs dropped by 60%+, and back-and-forth during testing was significantly lessened. It also freed me to move into the PM role I'd been building toward. Months later, after I'd fully moved off QA duties, the change still came up unprompted in anonymous 360 feedback from colleagues. The document became the team's standing QA philosophy, used to onboard every new QA hire after me.
Built Tebly, a year-end recap tool for tech bloggers combining AI-generated insights with developer humour and meme-based personas. Led development across a cross-functional team, shipped within a 10-day hackathon MVP scope.
Led the ideation process across all six team members, converging on the final concept through a cycle of divergence and refinement given the 10-day timeline. Users simply enter their blog link and receive AI-generated insights: a blog persona built from an LLM analysis of their posts, cringe-worthy AI-generated "quotes" from their own writing, and a "2024 in numbers" breakdown - all designed to be shareable.
Tebly won 1st place at the online hackathon "Potenday 412," voted on by fellow participants - mostly developers, our target audience. It went viral within "글도," a 600+ member tech-blogger community in Korea, with over 300 blogs and more than 17,000 posts analysed within a week.
Pro Baristas, a social enterprise, needed a new source of revenue. The opportunity wasn't obvious - it had to be uncovered directly from the people closest to the training programme.
Ran participatory research across trainees, operators and stakeholders to uncover an untapped revenue opportunity, then designed and iteratively validated Club Coffee through prototyping across multiple stakeholder groups.
The first prototype ran as a live session, structured around an aroma icebreaker, espresso and latte-art skill stations, and a timed mock café competition - designed to be both a genuine skills showcase and a fun, memorable experience.
The concept was adopted and is now in active development as a live service.
Venue owners and musicians faced friction in Capacity's entry experience, with two distinct user groups having different needs and expectations from the same onboarding flow.
Identified the root cause of friction in Capacity's entry experience through 35+ interviews and concierge testing across venue owners and musicians.
Prototyped and validated a simplified flow using AI-assisted tools, testing it hands-on before proposing it as the replacement for the existing onboarding.
The new flow replaced the existing onboarding. The project won the Social Impact prize.
Designing how the Justice Consultant programme - a paid role for people with lived experience of prison - could drive sustained engagement and meaningful participation within real-world constraints.
Low long-term engagement in a constrained, bi-monthly programme. Users - the Justice Consultants themselves - faced low trust, low continuity, and weak motivation signals across sessions.
Designed an engagement system balancing purpose, recognition, and agency. Key decisions:
Improved engagement depth, participation, and insight quality.