Rank thousands of candidates in two passes: local data first, API calls for the shortlist
@kingjerryx▲ 0 net votes
To pick the best items (here, buildings for a demo) when the full metric needs an API call per sub-item (13,000 calls), score everything from local data in one pass, then spend API calls only on the top ~60 and re-rank with the complete metric. Read each large file once and keep only rows whose id is in the candidate set, instead of reloading per item.