REC Philly is a creative coworking hub in Philadelphia. Studio bookings were handled by phone through a single front-desk coordinator, which caused overbooking, long waits and missed chances for creators to connect.

I designed the UX for a mobile app where members book rooms, pay and connect with other creatives. I worked within a brief the project lead had already defined, so requirements and scope were set before I joined. The information architecture, flows, wireframes, design system and high-fidelity prototype were mine.

REC Philly runs on memberships and room bookings, so bookings are revenue. The phone system created three problems that fed each other:

• Revenue leak. Rooms sat empty because members couldn't see or claim availability.
• Scaling ceiling. One coordinator set the limit on how many bookings the space could take.
• Churn risk. Double bookings and long waits wore down the trust members needed to renew.

I mapped the current and ideal experience to find the gaps, then prioritized the user stories with the team before any design began.

I designed the three flows members needed most, keeping each to a few steps.

I wireframed layout, content and flow logic first so I could test ideas with the team before polishing the UI.

The final UI carries REC's bold creative energy, kept simple on the high-traffic booking and membership screens.

I tested the prototype on site with Scarlett, who handled every booking by hand.

Rather than rely on remote feedback, I went to REC and walked through the prototype with her, watching where it fit her day and where it caused friction. That session changed two things a survey would have missed:

• She spotted unclear time slots straight away, so I made them easier to read.
• Her warm, helpful style became the "Scarlett" chatbot, so the app kept the tone members knew from the front desk.

Members can check availability, book and pay without anyone at the front desk.

The design removes the single-coordinator bottleneck. REC Philly's team reviewed and accepted the designs, and they were handed off for development.

I don't have data on how it performed after handoff. If I could measure it, I'd look at how many bookings happen without the front desk, how often rooms are double-booked, and how many empty hours get filled.

• Client meetings work best with a clear agenda and agreed goals.
• Visiting the space showed the real rhythm of bookings in a way remote feedback couldn't.
• One person's way of working can shape a feature. Scarlett's warmth became the chatbot.

The website was always out of date.

Woody Cat Rescue is a small, volunteer-run rescue. Cats arrived and got adopted every week, and workshops ran every month, but the website rarely changed. Only the founder could edit it, and she had no time.

An out-of-date site costs a rescue twice: people can't find the cats who need homes, and they don't trust a page that still lists cats who left months ago.

I interviewed the founder and mapped every way content reached the site. There were three, and each one depended on her.

• New cats. Volunteers already entered every cat in Shelterluv, their shelter system. Getting it onto the website meant doing it again by hand: open Wix, rewrite, upload photos, publish.
• Changes. A cat on hold, new photos, an updated FAQ. All of it needed a Wix login, so in practice nobody made these changes and the site drifted.
• Workshops. People paid through PayPal and picked a session in a Google Doc. The founder copied every signup into a Google Sheet and told people when a session was full.

I used AI only where the input is messy language, and plain rules everywhere else.

AI prepares the change. A person approves it.

Volunteers trust the website because it's right. So Claude never writes to the site directly: the bot shows what will change, from and to, with Save and Cancel buttons. Nothing goes live until someone taps Save, and every saved change is written to an audit log. If no cat matches, several do, or nothing would change, the bot says so and stops.

That one rule made it safe to let anyone on the team update the site from their phone, without a Wix login.

Meet the team where they already work.

Volunteers already used Shelterluv and Telegram every day. Instead of a new dashboard to learn, the system listens to the tools they had, so keeping the site current became a side effect of work they were already doing.

The founder estimates cats are now about 30% more likely to be adopted.

• The site stays current. New cats appear on their own, and any volunteer can update the rest from a chat.
• 3 to 5 hours a week back for the founder, who no longer retypes cats or tracks signups by hand.
• First donations. The site started bringing in donations for the first time: three donors so far, close to $1,000.

“Updating it is so convenient. I can do it right from my phone. And our cats are getting adopted really fast now.” Founder, Woody Cat Rescue

This started as a simple idea: take a written crochet pattern and turn it into a visual chart. It wasn't simple.

I'm a designer without a traditional engineering background, and this was my first time building a product end to end with AI. What I learned changed how I think about both design and building.

Designers sell crochet patterns on Ravelry, Etsy and their own sites. Charts are the format buyers prefer: clearer, faster to read and usable across languages. Drawing them by hand takes time, so many designers publish text only.

This tool lowers the cost of making a chart, so designers can offer the format their customers want. I'm testing with early users to see whether they'd pay for it.

Granny squares are one of crochet's most expressive formats, but their shape and colors often aren't clear until they've been made. To design one, you usually have to crochet it first.

My first idea was to have AI generate the visual straight from a prompt. Talking to experienced crocheters showed me why that fails: where a stitch goes depends on sequence and how the piece is worked, not on how it looks.

Those conversations also showed the real gap. Designers know charts are clearer, but making them takes time, so they skip it.

A crochet chart isn't just icons arranged on a canvas. A result can look visually correct and still be structurally wrong.

I started building fast in Cursor and it seemed to work, until the charts started breaking. I couldn't prompt my way out of it. The failure pointed to a decision I'd been putting off: what should a stitch actually be in this system?

Square and circular patterns follow different construction logic, so one placement model couldn't cover both. I wrote down the common failure cases and used them as checkpoints throughout testing.

A stitch is a logical unit, not an icon on a canvas.

The first version treated each stitch as a visual token. It was fast and looked right, but a token has no memory, rules or relationships, so editing and export kept breaking.

I stopped and rebuilt the data model: each stitch now carries its structure, placement rules and relationships, and the drawing is derived from that.

Users needed to adjust the result and still trust it. The chart and the written instructions both come from the same data, so an edit in one shows up in the other and they never drift apart.

681 people have tested it so far.

The app supports the whole workflow: paste a written pattern, get a chart, edit it by hand, and the instructions update. The structural approach holds up across pattern types. Built solo in one month.

• Niche creative workflows look simple on the surface. They rarely are.
• The real design work was deciding what the system treats as true, how much control users need, and how to make the result reliable enough to trust.
• AI speeds up execution. It doesn't replace understanding the domain or clear system logic.
• Learning to think in structures, not just visuals, was the shift that made everything else work.

Vertex Inc. makes tax compliance software. I was brought in to redesign three core partner pages to improve usability, update outdated content and increase partner engagement, so users can find the right partner while trusting what they see.

Desktop and mobile web, in Figma, with Crazy Egg and Google Analytics for research.

• Partner information was hard to find and filter.
• Outdated detail pages and broken links reduced trust.
• There was no clear call to action or benefit-led message.
• The pages felt disjointed and visually dated.
• Partner detail pages got little traffic or engagement.

To guide the redesign I combined analytics, competitor research and a content audit.

Analytics. Crazy Egg and Google Analytics showed most engagement on search and filters, very little on partner details, high bounce on first load, and no tracking on calls to action or card clicks. People wanted to filter but weren't finding what they needed.

Benchmarking. I studied more than 10 B2B partner pages, including Salesforce, Avalara and HubSpot. The strong ones lead with benefits and a call to action, use partner logos and testimonials, keep filters simple and cards consistent, and sometimes feature partners or tiers.

• The Partner Finder acted as the landing page but had no structure or introduction.
• Filters worked, but lacked grouping, clear labels and hierarchy.
• The layout couldn't scale as new partners were added.
• There was no narrative or visual rhythm to guide people.
• Users didn't know where to start, so discovery felt like effort.

Each change came from something the research showed:

• Clicks bunched on the top filters, so I regrouped the filter panel, gave it clearer labels and made it work on mobile.
• Users didn't know where to start, so tabs now split partners by type: Technology, Consulting or Integrators.
• The Partner Finder had no introduction, so the landing page now opens with benefit-led copy and a clear call to action.
• Detail pages got little engagement, so cards show only region, tools and a Learn more link.
• Outdated content and broken links hurt trust, so I restructured the content, fixed every link and added partner badges.

I worked closely with the UX manager and dev manager throughout, within Vertex's brand, technical and budget constraints.

• Shared wireframes and visual design early so the dev team could check feasibility.
• Delivered specs, naming conventions and layout guidelines in Figma.
• Gave the marketing team and web editors a content structure to work from.
• Built the layouts to be ready for multiple languages.
• Worked within technical and budget limits with the UX, development, content and web managers.

“The design looks so much cleaner and makes it much easier to search and find partners.”

Feedback from Vertex's internal team. The redesigned pages are live on Vertex's site.

I don't have a before-and-after number. The old pages had no tracking on calls to action or partner card clicks, so there was no baseline to compare against. Next time I'd set up that tracking before launch, so the redesign could be measured.

• Clear structure and hierarchy matter most in dense B2B content.
• Small changes, like clearer filters or stronger calls to action, can move engagement.
• Working with developers early avoids rework at handoff.
• Brand and technical limits pushed me toward clarity over decoration.
• Partner pages are part of the sales funnel, so they need to be designed to convert.