This looks pretty great! Do you have any sense of its performance? Very roughly, what kind of latency can one expect over, say, 1M, 10M, or 100M vectors? Ballpark: milliseconds, seconds, days?