Ari Nakamura
February 18, 2026 · 5 min read
Our new speaker detection model is 40% more accurate in overlapping speech and works reliably with up to 20 participants.
Ari Nakamura
Co-founder & CTO
Ari leads engineering at Nebula. Previously built transcription systems used by millions at a speech recognition startup.
Our original engine was good enough. But good enough was leaving accuracy on the table in exactly the moments that mattered most.
Most teams treat meeting notes as a minor inconvenience. The real cost is much larger than anyone admits.
Parallel replaced their daily standup with recorded async updates. Here is how they use Nebula to make it work.