Machine Learning

Overview

loading...

info

To edit this diagram, import the source file in draw.io, edit and export as xml again.

This picture depicts the current machine learning architecture in a high-level way. It shows the first recommendation service (for volunteering opportunities) and roughly depicts the flow of data throughout the architecture.

holi is just at the start of its ML journey and this architecture documentation is expected to see frequent updates.

In order to serve recommendations for volunteering opportunities, the following steps are run:

1. The Voltastics Crawler runs as part of a nightly Google Cloud Run job. It crawls all of betterplace.org and stores the results within BigQuery.
2. The data pipeline (source see holi-data-classification) then determines which of the crawled volunteering opportunities were new, updated or deleted.
3. For new and updated items, it runs a classification using Google Cloud AI services. The results are
   • a vector of similarities to all interests and skills a holi user can select for themselves in their profile settings.
   • a vector of similarities of the volunteering opportunity to other volunteering opportunities.
4. A last step in the data pipeline exports the volunteering opportunity items, along with the 2 classifications, into a Postgres database with the vector extension.
5. The volunteering service serves recommendations for users from the Postgres DB, based on either
   • the users interests and skills (personal recommendation), or
   • a given volunteering opportunity (similarity recommendation)