The biggest challenge was the long execution time. Originally, we used multiple parameters to compute similarity—title, description, cast, genre, region, etc.—but it took up to 20 minutes to return a response, so we implemented the following changes:
   •    Dramatically reduced the vectorization process (instead of combining multiple parameters, we merged the description and title and compute similarity on the cleaned, consolidated text only).
   •    Standardized movie descriptions to 300 characters, truncating any excess.
   •    Significantly reduced the number of matching genres (from 50 down to 30).​

finally solved