We found an embedding indexing bottleneck in JSON parsing

Nixiesearch: Running Lucene over S3, and why we're building a new search engine