sample draft / noindex
How I get useful in a new stack quickly
I often end up using tools I have never touched before. That happened with parts of my internship work, with TinyML on the ESP32, and with small utilities I write when a manual process starts taking too much time. I do not try to learn the whole stack before I start. I need enough context to make one real thing work and enough understanding to know where I am taking shortcuts.
I map the moving parts first
Before syntax, I want to know where state lives, what enters the system, what leaves it, which parts are slow, and what can break. A FastAPI service, an Electron app and an embedded inference loop look different in code, but this map gives me a useful starting point in all three.
Once I can describe the data flow in plain language, documentation becomes much easier to read. I know which sections matter now and which ones can wait.
I get one narrow path working
A small end-to-end result teaches me more than a folder full of notes. For a web service that might be one route that reaches the database. For an ML project it can be one sample moving through preprocessing and inference. On hardware, I would rather read a clean sensor value first and then add the rest of the system around it.
The first working slice gives me better questions. I can see the assumptions I made, profile the parts that matter and read source code with a purpose.
I leave enough context to come back
Fast learning becomes expensive when I have to rediscover it a month later. I keep setup commands, config notes, migrations, test cases and small decisions near the project. I also prefer scripts that can resume after a failure instead of forcing a full restart.
That habit matters more to me than remembering every API call. I can look syntax up again. Reconstructing why a system was built a certain way is slower.