Is it worthwhile to rebuild existing software for learning?

Be especially careful with security-sensitive features. Writing your own encryption, authentication, or mail-delivery protocols can teach you a lot, but those implementations shouldn’t go straight into production without serious study and testing. For an email system, also think about delivery when the recipient is offline, retries, queuing, and what happens when the two systems aren’t online at the same time.

There’s nothing wrong with simplifying the implementation if you’ve consciously simplified the requirements too. A Flask-based HTTP service might be fine for a small experiment, but it won’t automatically provide all the behavior and interoperability of established email protocols. Build it to learn, then compare its capabilities and limitations with the standard approach.

Not dumb at all. Rebuilding existing systems can be one of the best ways to understand how they work, especially when you’re doing it as a learning project. Comparing your version with established implementations afterward can also reveal design decisions and edge cases you hadn’t considered.

It depends on the goal. For practice, go ahead and build your own version. For a real product or work task, use mature libraries and services unless you have a specific reason not to. Existing solutions have usually been tested against years of unusual failures, compatibility problems, and security issues.