From plan to working product
A day of turning an English-practice product into a real foundation, with an app, domain logic, UI, Convex, dashboard, and agent coordination.
The day was about taking an idea out of inspiration mode and giving it structure. Instead of staying at the benchmark level, I moved into specification, tasks, technical foundation, and the first real screens.
English-practice product
The largest block of the day was a product for English speaking practice. I started by clarifying what the experience needed to do: guide a conversation, generate a useful report, and turn a session into clear feedback for someone building speaking confidence.
From there, the project gained an engineering base. The app was started with Next.js and Bun, then gained UI primitives, domain logic, date utilities, a landing page, a conversation report, and a data foundation with Convex.
I also aligned the interface choices. shadcn/ui became the component base, Motion became the natural path for animation, and Remotion stayed for later if the product needs a more mature video layer.
Agents working in parallel
I used parallel workspaces to split planning, implementation, review, and task closing. That required more organization than code: clear issues, small pull requests, closing comments, and contract reviews across domain, UI, and data.
I also reviewed the project skills. I removed overlap and kept the official Vercel skills as the reference when that made sense. The lesson was practical: the more agents run at the same time, the more each task needs a clear boundary.
Shareable calculator
In another product, I adjusted a plan calculator to accept URL parameters. A simulation can now open with the signup value and attendee count already filled in, while preserving the minimum fee and showing the result immediately.
This kind of change looks small, but it changes the sales use of the tool. Instead of asking someone to configure everything by hand, the link carries the scenario that needs to be discussed.
Public profile and career materials
I also worked on the public GitHub profile: clearer copy, metrics, featured repositories, and a better way to explain AI as engineering leverage, not low-level research.
In parallel, I reviewed career materials in a private and generic way. The focus was clarity: understand fit, prepare answers, and keep the portfolio coherent without turning this journal into a job-search record.
The result of the day was simple: ideas become more serious when they turn into small tasks, clear contracts, and a first version that can already open in the browser.