== Valui - Data science on AWS Lambda ==
A friend of mine was willing to create a real estate price predictor. I was
short of time and skills for this one. In order to make this project happen,
I recruited a few contractors.
One for develop and scrape the data. One for the data science part, who
developed a couple of models. One for develop the wireframes and layout.
Another to implement the backend and even other for the frontend. I was
responsible for managing all those people and to deploy models and code to
Aside all things, the most complicated part was to deploy the ML models to
AWS Lambda. AWS Lambda has a limitation of 50MB of code and the Python
libraries NumPy, pandas, SciPy and scikit-learn, plus their dependencies,
used to generate an over 250MB zip package. I had to learn how to minimize
and strip things to make everything fit under 50MB.
I succeeded and this allowed me to implement the models in a serverless
structure, which reduced costs, and, as the access was made through the AWS
api gateway, my friend/client was able to create a new product later on,
selling the usage of the models for 3rd parties.