This image-embedding based recommendation system is based on the following steps as charted:
By combining the name embeddings along with the existing image embeddings in a combined weighted cosine similarity score, we have more robust predictions.
A separate use-case, I've tried to work on: given a location and a search radius, bring out the top Tourist Spots according to preference (example 'Hills', 'Palaces' etc.)
However, Google Maps has this functionality. If you want to search 'Hills near me' on the Gmaps search bar, it will do it's job for you.
If you still want to check out, refer to the collab noteboo above.