What If Casting

Inspiration

Amazon Prime Video gives me streaming recommendations of what to watch, what if I spun that around. What if filmmakers were given recommendations to make better selections for who they cast, where they shoot, how much to spend on costumes? They could make better movies, right?

Originally focused on casting but Luminate Talent API was dropped from the hackathon, so now concentrating on any other recommendations for better film productions.

What it does

You tell it what genre and arena of film you're looking to produce, and an approximate budget that you have, and it recommends production choices made by similar, successful movies before in that same vein.

How we built it

It works based on a collaborative filtering recommender system over a sampling of up to 100 film productions similar to yours, and uses the k-nearest neighbors algorithm to identify the closest matches and suggest to you what you should do to make your movie just as successful as these others.

Challenges we ran into

Luminate Talent was removed from the competition, forcing us to pivot on what kinds of recommendations we could offer. This led to a shift to recommend below the fold talent like make-up artists, and filming locations used by previously popular films in the candidate genre and arena. Difficulties there included normalizing the data for variation in stage and screen titles (makeup artist, makeup assistant, makeup prosthetic artist, makeup teeth specialist, etc.); as a simplification we only selected talent whose titles literally include the text "makeup artist", but that needs to be made more flexible so as not to exclude anyone from being recommended.

API data exchange only offers small amounts of data, and ideally a RecSys works best with one of the bulk data exchange offerings like S3 import or Redshift shares. We get around this by selecting a randomized sample of film productions from Luminate's data sets that should statistically correspond to the film population as a whole.

Luminate Completed Films API would error with Wrong API Key so we created mock responses using the OpenAPI 3.0 response schema and some Web Search elbow grease to test minimal functionality when the API was offline.

To compare budgets apples-to-apples, we adjust budgets to 2022 dollars by assuming a 0.2% rate of inflation per month, calculated from the start date of a film's production. This would be a good opportunity to integrate an additional data API however, one with historic CPI readings.