June Care AI Host recommender

Inspiration

Right now, we have only rudimentary ways of recommending the best hosts to families. This improves that.

What it does

Recommends the best hosts to families based on data related to the host and family.

How we built it

Distilled data in our database using SQL into rows we could run a machine learning algorithm on. Then used sklearn to give each host a score given family information.

Challenges we ran into

Data amount and quality are lacking - we'll need to gather more data to do this really well. In particular it was quite difficult to find good labels for our data.

Accomplishments that we're proud of

We got the data and learning pipeline working end-to-end with reasonable latency and integration with the June Care API.

What we learned

We should think more about analytics and machine learning when creating our application schema. It would make future work like this much easier.

Also learned a ton about different machine learning algorithms like symbolic regression, as well as when it's appropriate to split datasets into training, testing, and validation sets.

What's next for June Care AI Host recommender

Next up we will add more features - particularly we would like to add some human feedback as labels as getting high quality labels was a major challenge.

Built With

• postgresql
• python
• ruby-on-rails
• sklearn