Side Projects

Built a web app that takes the nationalities of your team members and generates practical management suggestions based on Erin Meyer's The Culture Map. International teams are harder to manage than most people admit, and generic advice doesn't cut it when someone from Germany and someone from Brazil have fundamentally different expectations of what a direct conversation looks like.

• Users input the nationalities of each team member and their own nationality as a baseline
• The tool maps each person across Culture Map dimensions: communication style, hierarchy, trust-building, disagreement, and more
• Outputs specific, actionable suggestions on where friction is most likely to surface and how to navigate it

The most useful version of this wasn't the one that described the framework. It was the one that surfaced concrete tension points between specific pairs of nationalities. Specificity is what makes it actionable. Built entirely in Lovable in an afternoon.

A tool that lets you upload an ePub file and ask questions about the book, but only up to a chapter you specify. The problem it solves: you want to discuss or summarise what you've read so far without accidentally surfacing what happens next.

• Upload any ePub; the app parses it into chapters automatically
• Set a chapter cutoff and the tool only references content up to that point in its answers
• Ask anything: summary, themes, character motivations, what you might have missed

The chapter boundary constraint was the entire product. Without it, it's just a chatbot over a PDF. There are dozens of those. The constraint is what makes it useful for someone actively reading. Built in Lovable; the ePub parsing was the trickiest part to get right.

Built a custom GPT to help colleagues at BeZero draft their annual performance reviews. Most people find the blank page problem genuinely hard. They know what they've done, but translating it into structured, credible self-assessment is slow and uncomfortable.

• Trained on BeZero's competency framework and review format so outputs matched what reviewers actually expected
• Prompted to ask clarifying questions before generating, so it produced something specific, not generic
• Surfaced quickly and adopted informally across the team with no formal rollout

A narrow, well-scoped GPT beats a general one almost every time. The constraint makes it useful. The team adopted it because it saved real time on a task everyone dreads.

My local pilates studio releases class bookings at a fixed time each week. Popular slots fill in under two minutes. I was consistently missing them, so I automated the problem away.

• Built a script that monitors the booking system and registers for target classes the moment they open
• Runs on a schedule so it requires zero manual intervention on booking day
• Has a near-perfect booking success rate for classes that were previously impossible to get

The best automation is the kind you forget you built. I haven't manually booked a class in months. It also turned out to be a good excuse to get more comfortable with browser automation and scheduling infrastructure.