Christian Bolorinos

Post-event interviews kept surfacing the same three pains: unnavigable PDF calendars, inefficient networking, and a community that fragmented after every event. The Summit App fixed all three. Over 60% of attendees used it, and it continues getting engagement up to two months after events.

Context

WSA is a UN-backed organization that gives awards to projects supporting the UN Sustainable Development Goals. Several times a year, they host regional and global events to celebrate award winners and provide valuable networking opportunities for attendees. I was in charge of the WSA website (there is another case study about that). To make the website more helpful, I often followed up with 5–10 people I had met at each event. I asked them what they were hoping to get out of the event and what could be improved.

Problem

In these interviews, the same themes kept coming up. One was that the calendar (a PDF sent through a WhatsApp group) was hard to navigate. Another was the inefficiency of the networking. A person working in agritech in Japan had very little to say to someone working in fintech in Chile. Also, the jumble of emails and LinkedIn connections attendees ended up with was often difficult to sort through and follow up on. The WSA community would come together for these events, but would mostly fragment again immediately afterwards.

What I did

My first impulse was to address these issues through the website. I built several MVPs, including an AI agent that had access to attendee and project data and suggested connections aligned with each user’s goals. These solutions received mostly positive feedback but were clunky and often drowned out by other communication channels. I also evaluated several SaaS solutions. While many solved the calendar problem, none effectively improved networking or helped maintain community engagement after the event. I considered building a custom app, but WSA did not have an in-house mobile developer or the budget to hire an agency.

Team and constraints

Summit was built by two people for one very clear stakeholder. Emeka, a developer I met through the WSA community, built the app and worked partly for equity. Making it white-label, so it could serve clients beyond WSA, was his idea, and it’s the reason Summit is an independent product today. Peter Bruck, WSA’s Chairman, was only interested in an app if it kept people engaged between events and strengthened what WSA really offers its members: network capital. That set the bar for every decision. A better calendar alone wouldn’t meet it, so the calendar was only the first iteration. The second focused on projects, the community feed and connections.

The Summit App

Explore it on the Summit website.

The first iteration of the app focused solely on calendar functionality. It provided a clear schedule of all sessions, allowed users to sync events with their personal calendars (iCal, Outlook, or Google), view session locations via Google Maps or Apple Maps, and access speaker profiles and LinkedIn pages. The app was embedded within the website so that users who preferred not to download an app could still access the schedule. This successfully solved the calendar problem but did not yet address networking. However, it helped onboard users, which made later iterations easier to roll out.

The second iteration retained the calendar and introduced two new sections: Projects and Community Feed. The Projects section displayed profiles of all winning teams, their objectives, and their social links, allowing attendees to follow up after the event. The Community Feed functioned as a forum where participants could post what they were looking for (investors, employees, or partnerships) and discover others with similar goals. The app also allowed users to browse attendee profiles and identify who might be most relevant to connect with.

The app also featured an AI agent that could provide information about winners, the agenda, and attendees. This agent supported networking during the event by suggesting people to connect with based on each user’s expertise and objectives. It also helped attendees use their time more efficiently by recommending the most relevant sessions.

We expected the app to attract spam if it got into the wrong hands, so we handled access during onboarding. Admins upload a CSV of their mailing list, and anyone who signs up with an email on that list gets full access. Everyone else gets guest access, which hides the forum and the attendee directory, unless they have an invite code. Codes went out in email campaigns and on QR codes at events, and expired after a set time.

Because the app was white-label, we had to decide which parts of the brand clients could customize (colors, fonts, logo, some of the wording, and which features were switched on) and which stayed fixed (the glassy textures, the gradients and the core functionality). The design library was built around that split, and so was the code: one theme file holds everything customizable, so swapping it changes the whole look of the app.

Results

That second number is the one Peter cared about: people were still using the app to follow up long after the event ended.

Summit has since spun out of WSA and now operates as an independent product, serving clients with no connection to the organisation it was originally built for — among them Gomde and Quinta del Sordo.

Related WSA work

This grew out of a wider remit at WSA that also covered the website and the identity for their flagship event.

See the WSA Country Pages case study →
See the WSA Global Congress Branding case study →

Applicants had to track down their country’s local expert just to learn how to apply. I designed and built dynamic country pages that centralized everything in one place. Within three months, 19% of all users were visiting them and application volume rose significantly.

Context

WSA is a global initiative supporting digital projects aligned with the UN Sustainable Development Goals. Because any UN member state can participate, the organization operates through a distributed network of local coordinators who manage country-specific application processes and community engagement.

Problem

On the previous website, this created significant friction for applicants. Users had to manually find their country’s local expert and contact them to understand the application process, which varied widely between regions. As a result, potential applicants often struggled to understand how to participate or how to engage with their local WSA community.

What I did

To address this, I designed and developed a system of dynamic country pages that centralized all local information in one place. Each page displayed the country-specific application process, linked directly to the correct submission database, and showcased local winners, youth ambassadors, national experts, and upcoming events. The pages followed a standard treatment to stay consistent and maintainable globally, but were also customizable by the local experts to reflect each country’s process. This allowed users to quickly understand how to apply and provided clear entry points into their local WSA network.

Team and constraints

There was no in-house development team. Stefan Rath, the developer who had built the original WSA site, was on a four-hours-a-month contract, so I designed the pages and wrote most of the code myself. Stefan stepped in on bugs I couldn’t crack, and helped me connect the pages to a country ID that already existed in the database, so national experts and projects would be linked to the right country automatically. I worked with WSA’s community manager to map how the application process differed between countries, so the system reflected local processes while staying maintainable across every country.

Results Beyond the country pages

In addition to the country pages, I worked on improving regional engagement through WSA’s newsletter. Using location data available in Brevo, I segmented WSA’s 20,000-contact mailing list into five geographic regions. This allowed us to promote local events and highlight regionally relevant success stories, which led to improved engagement and higher open rates in subsequent campaigns.

I also produced an explainer video for WSA, creating motion graphics that helped communicate the organization’s mission and participation model more clearly to new audiences.

Related WSA work

The interviews I ran to improve this website are what surfaced the problems behind the Summit App, and the same remit covered the identity for WSA’s flagship event.

See the Summit App case study →
See the WSA Global Congress Branding case study →

Ria’s mobile app had to launch across five markets at once, in the middle of a pandemic. I was in charge of enforcing the newly defined brand guidelines and doing some custom illustrations and animations. I also organized the feedback sessions between product and customer support that helped inform product direction.

Context

Before I was involved in the Ria app, I had spent the better part of a year building the internal infrastructure for Ria’s visual identity. This included a Digital Asset Management system that gave every team (developers, designers, comms, marketing, and brick-and-mortar operations) a single source of truth for brand assets. It also included an internal intranet and a comprehensive set of brand guidelines built in Frontify, which formalized Ria’s evolving visual language across touchpoints for the first time.

Problem

That groundwork is what got me looped into the app. Ria was preparing to launch its mobile app across multiple markets simultaneously, and the design team needed someone who understood the brand deeply enough to ensure the app’s visual language felt continuous with everything else Ria was putting into the world. The design challenge was one of translation. The warmth and cultural specificity of Ria’s social media presence, the visual codes of the physical locations, the illustration style we’d developed for marketing; all of it needed to find its equivalent in a mobile interface, which operates under very different constraints.

What I did

My contribution had a few distinct threads. I worked directly with the product team to introduce the illustration assets I had developed into the app, reformatting them for mobile dimensions and animating a selection in Lottie to guide users through key flows and improve comprehension at moments of friction. I also ran sessions showing the product team how to use the intranet and brand guidelines correctly.

Beyond the brief

The other thing I did wasn’t in my job description at all. Ria’s product team was based in Colorado. Customer support was in Madrid, which was where I worked. The two departments had almost no structured communication, and as a result product decisions were being made without a clear line of sight into the issues customers were actually experiencing. Having studied UX, I knew what that gap cost in terms of wasted iteration cycles and avoidable friction. So I organized the first monthly cross-department feedback meetings between the two teams. I ran the first sessions, and handed it off once it was established. The app launched as a critical continuity channel during the early months of the pandemic, when Ria’s physical network was severely disrupted, and those feedback loops proved essential to identifying and resolving friction points quickly under pressure.

Results

To see the full app you can download it on Android or iOS.

Related work

The brand language the app inherited came out of the corporate communications work I was doing for the same client. See the Ria Corporate Communications case study →

As 5Mins scaled from less than 1k to more than 7k monthly users, repeat usage plateaued after onboarding. I was responsible for redesigning the engagement layer of the product, focusing on social interaction and gamification to encourage sustained learning behavior. My work increased 7-day retention by 18% and improved repeat usage across the platform.

Problem

When I joined, the home page primarily surfaced content automatically based on users’ skill sets. While this worked at smaller scale, it offered little visibility into team activity or shared learning.

What I did: social home experience

I redesigned the home experience to introduce a social feed inspired by familiar interaction patterns from TikTok and Instagram. The goal was to leverage interfaces users already understood, reducing the learning curve and allowing them to seamlessly browse, share, and discover learning content in a way that felt natural.

This included introducing a Share feature that allowed users to distribute videos within or outside the app, as well as a social feed that surfaced content suggested by coworkers, trending within teams, or assigned by managers. The decision to prioritize social features was also informed by broader platform history. As described by Tim Wu in his analysis of early internet platforms, services like AOL gained dominance not through content, but by embedding social interaction directly into the user experience. Applying this principle to 5Mins, we focused on making learning activity visible and shareable within teams, rather than treating it as an isolated task.

What I did: gamification system

The platform initially included basic streaks and daily goals, which proved insufficient for long-term motivation. I designed a structured learning path supported by badges and certificates, drawing inspiration from Duolingo’s progression and reward systems.

Users unlocked badges by earning skill points through watching relevant videos. After unlocking three badges, they became eligible to earn certificates. Upon completing the required lessons, users were prompted to take an assessment. Since every user watched different videos, there was no way to automatically generate quizzes, as each learning path was unique. To solve this, we included an AI agent powered by ChatGPT. It analyzed their recent lessons and generated quiz questions. Users scoring above 80% received a shareable certificate, meaning they could demonstrate tangible proof of their learnings on LinkedIn.

Team and constraints

I reported to the PM, whose priority was the KPIs shown to investors. The company depended on its next funding round, so retention and engagement numbers were the brief. The developers were overstretched, so every design had to move those numbers without adding much build time.

That made the design system a constant negotiation between two sources of truth: my Figma library and the developers’ Storybook. Rather than forcing one to always win, I reconciled them component by component. Where Storybook had drifted on brand consistency or information hierarchy (inconsistent CTAs, mixed font usage, mismatched corner radii), we brought Storybook back in line with Figma, because those inconsistencies erode trust in the interface. Where the developers’ version of a component simplified the app’s logic, I adapted Figma to Storybook. Design and code stayed in sync, and the developers didn’t have to rebuild anything that already worked.

Results

At SwissBorg, I led initiatives to optimize the marketing website, improve app downloads, and validate new product offerings through UX research and experimentation. Over six months, the website’s conversion rate doubled.

Context

I joined SwissBorg on a 6 month contract to work on conversion, and everything below was designed, tested and shipped inside that window.

I had a lot of autonomy. The PM’s goals were simple: more app downloads and more premium upsells, and how to get there was up to me. I proposed running A/B tests; the developers already had the infrastructure, so it was easy to start. I also set up Hotjar, so we could see which interactive components people actually used, not just which variant won.

What I did

The website’s main goal was to drive app downloads. I designed and executed conversion optimization initiatives. One of these was running continuous A/B tests, and prioritizing high-performing components. One clear example was the QR code on the home screen, which was originally visible only on hover. Showing it by default was one of the changes behind the doubling of the site’s conversion rate.

Beyond the website itself, I wanted to know which of SwissBorg’s services converted best, so I asked. Nobody had an answer. So I designed a social media campaign where each ad featured a different service while every other variable was held constant: video length, time of day, the template used to make the videos, and so on. The results informed prioritization of offerings and contributed to a 25% reduction in social media marketing costs for the same results. It also showed us which services needed to be featured more prominently on the website.

I also optimized the premium upsell experience on the website. Using interactive yield comparison graphs, we highlighted subscription plan differences and token-based rewards (CHSB) while maintaining clarity for users of varying expertise. Hotjar data helped identify which interactive components drove conversions, and an accordion pattern allowed users to access complex information without cluttering the main experience.

Results

This project focused on redesigning the website for Barclay Damon, an international law firm with 9 offices, 65 practice areas, and hundreds of attorneys. The core challenge was not visual refresh, but scale: creating a system that could accommodate a large, complex organization without fragmenting the brand or overwhelming users.

What I did

Early in the process, I explored multiple strategic directions for how the firm could present itself digitally. One concept emphasized Barclay Damon’s Renewable Energy practice, while another used a more abstract, flexible visual system capable of representing a wide range of legal domains. These explorations were used to align internal stakeholders around what the website needed to do for the business.

The final direction prioritized adaptability and longevity over expressiveness. The website was designed as a modular system, allowing hundreds of pages to be created, maintained, and evolved without design conflicts. Motion and visual differentiation were used selectively to avoid compromising clarity or performance.

Team and constraints

At Saturno I worked between a PM from a design background, who pushed for beautiful, current, craft-driven work, and a small dev team who pushed back for simplicity and reduced dev time.

Results

While my role included hands-on design, the strategic value of the project lay in translating organizational complexity into a coherent digital structure that balanced marketing needs, editorial autonomy, and long-term maintainability within a custom CMS environment.

Most people knew Ria as a small high-street storefront where you went to send cash home. Useful, but a long way from how anyone pictured a technology company. That perception was Ria’s brand problem in a nutshell.
Meanwhile, a new generation of fintech darlings (Wise, Revolut, Remitly) were building glossy digital identities and becoming household names in tech circles, despite serving a fraction of the markets Ria quietly dominated.

Problem

The truth was quite the opposite. Ria was one of the largest money transfer operators in the world, moving billions of dollars a year to countries and corridors that Wise had never even considered. It was a lifeline for migrant families, a financial artery for communities in parts of Africa, Latin America, and Southeast Asia where most people had never heard of Revolut. My department’s mandate wasn’t marketing in the conventional sense. It was to change how the world saw Ria, starting with prospective talent and potential partners, and to make visible a story the company had never bothered to tell.

What I did

The strategic anchor was repositioning Ria as the fintech that served the bottom billion. This meant moving away from transactional messaging and toward storytelling that foregrounded the human and economic weight of what Ria actually did. One of the most effective executions of this was the Orange Envelope. This video followed the emotional journey of a remittance payment from a worker abroad to their family at home. It showed what the money meant to the recipients and how it could change their lives. On the other end of the spectrum, our Microfinance video educated audiences on a corner of financial systems that mainstream fintech commentary almost entirely ignores, and one where Ria had more operational depth and nuance than any of its better-known competitors.

Content was distributed across Instagram, LinkedIn, and YouTube, with topic selection driven by keyword research: we targeted concepts that were frequently searched but underrepresented in quality content. Measuring success, however, required more precision than follower counts or likes. We defined specific KPIs around LinkedIn engagement from decision-makers and senior roles at top fintech companies to track whether the people we actually wanted to reach were interacting with our content. This distinction mattered because any post mentioning Atlético de Madrid, for instance, would generate enormous engagement from football fans. Unfortunately, these posts almost never got engagement from the audience we were trying to influence. Filtering for role-based engagement gave us a much cleaner signal about what was resonating with fintech leaders specifically, and let us iterate content strategy based on data that was actually meaningful to our objectives.

Results

My role spanned research, scriptwriting, animation, and production, but what I’m most proud of is the strategic coherence of the work. Through videos on immigration, finance and the impact of remittances we were able to reposition Ria as an important player in the fintech ecosystem.

Related work

That same visual language later had to be translated into Ria’s mobile app for a five-market launch. See the Ria Interface Design case study →

Immersive research uncovered that couriers’ biggest blocker was Spain’s freelancer paperwork. Addressing that helped to grow the courier base 7x and won contracts with Glovo, Uber Eats, Stuart and Getir.

Context

I was hired for a 6 month contract at Service Club Delivery, and while my formal role was as Creative Director, my informal roles included Web Developer, UX/UI Designer, Social Media Manager, UX Researcher and Motion Graphics Designer.

The company was formed to connect couriers to courier companies (Uber Eats, Glovo, Deliveroo, etc), and to reduce their churn rate (which is one of the biggest pain points for these companies). Our services included a job board and an academy where couriers could upskill and earn certificates (think LinkedIn for couriers).

Problem

When I joined SCD I also signed up to be a courier for Deliveroo to better understand the ecosystem, riding around Madrid delivering food for about 3 hours a week. I also interviewed countless couriers, both in person and online through surveys, interviews and data analysis. I soon found the massive pain point that they had, and a way we could fix it.
The hardest part about working for the delivery apps was the infernal pile of indecipherable documents that one had to complete to be able to register as a freelancer. This was made significantly worse by the fact that most couriers are either students or migrants.

What I did

The first step we made was to add a course in the academy on getting and maintaining a Spanish work visa, as well as several others helping people understand what work opportunities were and weren’t available to them. Users had to create a profile on our system to access these courses, which were marketed extensively on online communities that were popular with students and recent migrants. Additionally, we partnered with a company that registered people as freelancers for a monthly fee of 12 euros. This meant that through our interface we were able to offer people a way to fill out this immensely complicated paperwork without lifting a finger.

Results What I learned

The insight that transformed the business came from outside the product. Riding the bike surfaced what surveys alone never would have.

A chatbot I built to stop answering friends’ visa questions became a self-sustaining relocation business. AI handles the education and qualification; humans handle the high-stakes legal work.

How it started

Nomads Relocation started as a WhatsApp message. A few friends were planning to move to Spain and kept asking me the same questions about the Digital Nomad Visa, about tax registration, about what paperwork they needed. I built a simple chatbot to handle those questions so I didn’t have to answer them manually every time.

The chatbot was more useful than I expected. When a friend who worked as a real estate agent in the US mentioned that many of her clients relocating to Spain were running into the same friction, I embedded the chatbot in a proper website and started paying attention to the conversations people had with it. The pattern was consistent: once someone had figured out their visa eligibility, they generally asked for an immigration lawyer. A small portion asked about housing, banking and insurance as well. Each service we added came directly from questions the chatbot was already fielding.

How it works

The customer journey now works like this. Most users find us organically, but we also go looking for them. Automated workflows monitor Reddit and Facebook for people mentioning a move to Spain, asking about the Digital Nomad Visa, or expressing frustration with the paperwork process. An AI agent assesses each post for relevance and drafts a response; I review and post it manually, since both platforms penalize automated behavior heavily. The post leads them to the website where they can interact with the chatbot. Around 5% of the people who talk to the chatbot go on to pay for an intro call with an immigration lawyer. The chatbot costs almost nothing to run, and the conversations it has before people are ready to commit build trust. By the time someone does convert, they’ve already been educated on the process and pre-qualified themselves. The low-touch freemium model also means the business requires very little of my time on a day-to-day basis, which was always the intention.

Results What I learned

Building Nomads taught me something I’ve since applied to other projects: the combination of AI and automation is incredibly powerful and can make processes much more efficient. The business itself only exists because I was already automating conversations. Once I understood what that made possible, I started automating the sales funnel the same way (monitoring Reddit and Facebook for people mentioning a move to Spain). What started as a tool for helping a handful of friends became a small but self-sustaining business built almost entirely on automation.

During my Master’s in Customer Experience & Innovation at IE, I was part of a small team commissioned by Hilti to investigate the drivers behind a high churn rate among lower-income customers. The problem was framed not as a usability issue, but as a mismatch between Hilti’s value proposition and the financial realities of this segment.

Research

We began with large-scale qualitative research, analyzing over 4,000 customer emails to identify recurring themes. From this dataset, approximately 200 customers completed structured questionnaires, and around 20 participated in in-depth interviews. This allowed us to move from anecdotal feedback to validated behavioral patterns.

Findings

Our research revealed that churn was not primarily driven by dissatisfaction with product quality, but by poor understanding of long-term cost structures. Many customers were making purchasing decisions without a clear picture of when renting versus buying made financial sense, leading to regret and disengagement.

Based on these insights, we mapped the customer ecosystem and end-to-end journey for this segment, identifying key decision points where Hilti could intervene earlier. Rather than proposing major changes to the product line, we recommended targeted adjustments to how value was communicated.

The intervention

Our final proposal combined strategic insights with a concrete intervention: a value calculator designed to help customers compare rental and purchase scenarios based on their usage patterns. The goal was to enable more informed decisions upfront.

Rangjung Yeshe Gomde is a Tibetan Buddhist retreat center in the Austrian prealps. They are a long-standing client and one that reflects my commitment to aligning my work with projects that have a genuinely positive impact. I designed and built their website from scratch, but what I ended up delivering was considerably more than a website.

The brief

The original brief was straightforward: create a clear, bilingual platform for announcing retreats and events, with a booking system that could handle extended stays. Gomde’s audience is roughly 75% German-speaking and 25% English-speaking, so the multilingual setup was a core requirement from the start, as was making the site maintainable by non-technical staff.

Problem

The automation work grew out of my conversations with the staff, many of whom are close personal friends. While I was setting up the booking functionality, I started asking the staff about how they currently managed registrations. The answer was printed calendars and spreadsheets. Every retreat required manual data entry across multiple systems, and attendance information, dietary requirements, and accommodation preferences had to be tracked and communicated by hand. The more I understood the workflow, the more obvious it became that almost all of it could be automated with n8n.

What I built

The system I built works like this: when a staff member creates a retreat or event, a registration form is generated automatically, with dates pre-populated from the event but editable by participants. On submission, attendees receive a confirmation email, are added to the event as Google Calendar attendees, and are enrolled in an automated pre-arrival and post-departure communication sequence. Staff can see at a glance (in Google Calendar) how many people are coming, when, what accommodation they’ve requested, and any dietary requirements. What previously required hours of manual coordination now takes almost no time. Participants are also synced to Mailchimp with language and interest-based tags, replacing the previous approach of sending everyone the same communication regardless of language or program interest.

Beyond the website

I also produce Gomde’s monthly podcast, which averages around 800 listeners per episode.

Related work

This work led to a second project in the same community: a site for what will be the first Buddhist university in Austria, with an AI editor its founders can drive themselves. See the Kremsegg University case study →

Event planning was new for me when I first joined WSA. But I loved (and still love) the project. Thousands of impact entrepreneurs from all 182 UN Member States getting together to connect, to learn from each other, and to help bring to fruition the UN Sustainable Development Goals. Imagine having lunch with someone who was designing an AI that blocked hate speech, then coffee with someone who was working on an app that helped people navigate their pilgrimage to Mecca, then getting the cab back to the hotel with a crew that was helping disabled people travel Australia. I feel extremely privileged to have been part of this.

Like most idealistic ventures, funding was tight and the team had to wear many hats. My role included building the digital infrastructure, coding customizable country pages for each of the 182 UN member states, building an app that helped people navigate the events with AI-powered networking suggestions, and, my personal favorite, branding the events.

I’ve always enjoyed branding, and this brief combined it with travel. Each Global Congress identity brings together the host city, the themes of the event and sometimes its co-hosts. The one shown here is for the 2026 Global Congress in Vienna, drawn from the city’s landmarks and the banks of the Danube, with the sharp geometry of our co-hosts, Vienna Up, worked in. The brand lent itself well to animation: sponsor logos were built into looping animations of Viennese landscapes, designed to catch the eye between sessions without pulling attention from the speakers. The aim was the mild pleasure of a walk through Prater Park.

The stage elements included a set of cubes: low-budget objects that make a stage feel branded. Arranged in different ways, they could show the Danube, the Prater, Vienna City Hall or the sponsor logos, depending on what each session called for.

WSA’s Chairman is a big believer in print and wanted a booklet with the winners, sponsors and background on the Global Congress. I suspected most copies would end up in a landfill, so I added note-taking exercises for the pitches, workshops and panels to give people a reason to keep it. I have no data on how many booklets were thrown away, but I did see a lot of people filling in the exercises.

Related WSA work

The other two hats mentioned above — the country pages and the event app — each became a project in their own right.

See the WSA Country Pages case study →
See the Summit App case study →

An early-stage university needed a simple site that its non-technical founders could keep changing as the project evolved. So alongside the website, I built a password-protected AI editor that lets the client edit any page in plain English, preview the change and publish it, with every version saved to GitHub.

The context

Kremsegg University aims to be the first Buddhist university in Austria. It’s still early days: the founders have the building and are raising the funds they need to secure permits. One of the founders asked me to build their website. The brief was modest: a calling card, a place where people interested in the project could learn more, and donate or invest if they felt inclined.

The problem

I built the website on Claude. I’d been using Claude to write front-end code for a while and found it reliable, so the build itself went quickly: a donate page, a contact page, and everything else a landing page needs. But the project was moving fast. New donors wanted revisions (wording, photos, small details) and I found myself making constant updates that each took under 30 seconds to implement but broke my flow every time. It was inefficient for everyone. Worse, I realised I was essentially just copy-pasting the client’s requests into Claude and pushing the result.

The solution

That last realisation is why this otherwise ordinary website is in my portfolio. If the client’s requests could go straight into Claude, the client could send them there themselves. What they needed was a safe way to do it.

I added a password-protected AI editor to the site. The client picks the page they want to change, describes the edit in plain English, and can upload images. Claude, via the API, applies the edit to the site’s code and shows a preview. Nothing publishes until the client confirms, and every change is committed to GitHub first, so any mistake is one step from being undone.

Most of the design work went into what the editor can’t do. It only responds to requests about the website, so it can’t drift into general chat. The client chooses the page before writing a request, so the model only ever sees that one page rather than the whole site. And if a request is too big to handle safely, the editor offers to email me instead of attempting it. I’ve found AI tools work best inside tight limits, and here the limits sit on the model, not on what the client can ask for.

In other words, the client can now edit their own website just by describing what they want, tailoring it to a shifting audience without friction. They were delighted. I won’t pretend the experience didn’t unsettle me a little. I’d essentially designed myself out of the equation. But I’ve come to believe that evolving your value proposition alongside AI is the only way to avoid being displaced by it.

Related work

This project came through the same Austrian Buddhist community as the retreat center whose site and booking automation I built. See the Gomde EU Website case study →