Woo hoo! I made it!
I can’t believe this documentation takes more time than the project itself lol
Although the time spent for this project alone is around half a year, some components were built more than three years ago. It’s basically a project combining my academia ML modeling knowledge with the backend and ML Ops skills I acquired from the industry as a ML Engineer. It’s been an awesome journey so far. 
This project started with an idea during my shower (and yes, I would proudly admit 80% of my ideas came out while I was taking a shower instead of working or studying…)
It started with building a ML model service with REST API. After finishing it on my Linux environment, I found there were issues when I tried to deploy it on my Mac laptop, and therefore I turned to Docker to make my life easier.
Since we were using Docker anyway, why not try GPU-enabled image and the side project Machine Translation with Transformer came in handy.
After I included the machine translation model in this project, I got two ML model services. It’s time to play with the microservice architecture. I built a repository including multiple ML models and the goal is making the deployment and maintenance as easy as possible with Makefile.
It’s cool to make it public and have family and friends test it. I started to wonder about the qps that my desktop could handle. I found JMeter could be a great tool for me.
Finally, everything looks good on the backend side. However, when I tried to have my wife test it through the terminal, she frowned at the black screen with some white text jumping around. All right, I got it, let’s build a more user-friendly interface. I built a crappy frontend interface allowing users to interact with the service through browsers.
I still have a lot of items I want to try. The best part as an ML Engineer is that we would need to take care of both backend and modeling stuff. I would love to see how I could build a recommendation service with some streaming processing techniques and things like Feature Store. Anyway, let’s not stop here. Any comments or feedback are welcome! However, since it’s a learning-based project, the reason why a weird component is there would probably be I just want to learn how to use it. 
Keep the ball rolling!