MACHINE LEARNING
Python is the default go-to for machine learning. It has a wide variety of useful packages such as pandas and numpy to aid with ML, as well as deep-learning frameworks. Furthermore, it is more production-friendly compared to other ML languages such as R.
Pytorch is a deep-learning framework that is both flexible and fast compared to Tensorflow + Keras. It is also well documented and has a large community to answer lingering questions.
Dev Tools
IDE: VSCode is a lightweight IDE that supports multiple extensions that can help our development process. For example, linting extensions such as prettier, or local SQL server extensions.
Version Control: Github is a widespread version controlling system. Our developers are familiar with it, and Github provides many features to increase productivity, such as Github Issues.
CI: Github Actions are an integrated CI/CD tool provided by Github. This makes it the simplest and most straightforward choice for CI.
Hosting
We are going to host our MVP app on Heroku. Heroku is developer-friendly and provides generous services on their free tier. Additionally, there are many tutorials on how to deploy to Heroku.
We are planning to use AWS Lambda for ML inference. In our MVP workflow, we spin up a server, load model weights from S3, read data from our database, perform inference, write those predictions into the database, and close the server. This is a perfect use case for AWS lambda.
S3 is our service of choice for storage. We are going to store model weights (or models stored as pickles).
