Tampus Cafe: A Fun Way to Focus

Tampus Café —a play on tempus (time in latin) & an espresso tamper— is essentially a pomodoro timer wrapped in a turn-based coffee shop business strategy game. I’m a big fan of the pomodoro technique, but there’s nothing particularly inspiring me to use it on the daily. So I asked myself: how might I encourage and reinforce a pomodoro habit by employing some cozy game magic into the mix?

The Insight

As a 90s/2000s internet kid, I have never really stopped thinking about the game Lemonade Stand (and later, Dope Wars). I’ve tried to rebuild it numerous times as a side project (with hand-coded JS and Redux stores) but each time realized that business modelling & calculus were not really how I wanted to spend my weekends.

Enter 2024 and the birth of '“vibe coding.” I immediately tried to get the job done multiple ways using multiple tools to limited success (the state of LLM hallucinations was a bit too extreme at the time, and I’d end up with something that looked cool, but the gameplay logic was spaghetti underneath. 

Over the past 2 years, as we’ve tried and tested a variety of ai-prototyping and engineering approaches at work (memory banks, context engineering, multi-agent-workflows (not a fan), etc etc), I started seeing some utility in the idea of recursive learning (training agents to update and “teach” themselves about sticking to the plan, debugging, and creating better records/context for themselves.) Then one day on github, I noticed a trusted dev friend follow a repo called /compound-engineering (similar to Superpowers). It is essentially a Claude Code plugin that has codified, refined, and provided a comprehensive, well-reputed, and nicely orchestrated set of observable workflows and agents that felt like what I’d been dreaming of.

As someone who is happy to not-reinvent-the-wheel, I figured this personal project might be complex but comprehensible enough to test the plugin on. So far, it has been a delight to use, and really adds a layer of depth, persistence and consistency in agent behaviour that felt like the biggest challenge to uphold. There’s lots of great tools for helping form awesome PRDs and plans, and though native “memory function” has improved significantly over time, it’s been a huge win for improving the consistency of session-to-session (and project-specific) behaviour through learning.

All that to say: it’s still really important to watch your code, take regular pauses to regroup and rebase, and know what’s what. While I can’t take full credit for the calculus that Claude came up with for the game engine, I did know to ask to explain it to me at the right level of technical detail so that I could understand and learn how the gameplay engine and elements are all interconnected.

While this is just a fun personal side project, I hope this project can showcase my proficiency with these tools, and enthusiasm for this new way of working. AI accelerated engineering tooling is incredibly powerful, but does need to be approached with caution and care.

All of this, including the graphics, code, gameplay engine, additional SME agent/skill development and accessory features like a dev panel and furniture placement UI were spec’d and generated in approximately 30 hours of active work.

Current Status

This project is currently under development with plenty of features to come. Up next is onboarding and gameflow refinement, with audio integration (SFX, focus music, and alert sounds), website/app blocking during focus, notifications, and additional game features like a furniture shop mini-game, staffing, and recipe builder.