I've been experimenting with a few ideas to make my ML workflow more re-usable:

Exp 1: http://amitness.com/toolbox 
- Track ML libraries/tools/services in markdown organized by phases
- Previously: Cluttered bookmarks and hard to find
- Now: Easily access useful tools for "task X"

Exp 2: http://amitness.com/notebooks 
- Personal notebook snippets on how to do X in library Y
- Idea is to document runnable minimal code for task X on a small dummy data for reference in future
- I use it along with http://notebooks.quantumstat.com  by @Quantum_Stat

Exp 3: http://github.com/amitness/learning
- A learning-in-public log
- Idea is to do collect resources and do bulk learning for one theme at a time e.g. say "pytorch lightning"
- Easy to refer to resources that worked for me when someone asks "How to learn XYZ"

Exp 4: http://amitness.com/cookbook 
- Organized dump of commands/setup process I need to refer a lot (e.g. nginx, ssh flags etc.)

Exp 5: fns (wip)
I'm also writing a python library for functions that I have repeatedly need.

I started thinking on this inspired by @nlpguy_ ( http://models.pratik.ai/ , http://deployment.pratik.ai , http://ask.pratik.ai/ ) and @fishnets88 who builds small useful libraries to solve problems he faces.