Careers · AI · San Francisco · In Person
Founding Engineer, AI
Speaker identification that abstains instead of guessing.Cross-session re-identification is validated. Your job is to make it compound.
What you will do
- Own the speaker identification and enrollment pipeline on the Mac, and later on the phone.
- Tune for precision over recall. A wrong attribution is worse than an unresolved one.
- Turn user corrections into enrollment updates so recognition improves with use.
- Measure how recognition decays across days and beat it.
- Detect names from the conversation itself.
What we look for
- Hands-on work with speaker embeddings, diarization, or adjacent audio ML.
- Comfort shipping models that run locally on Apple silicon.
- You care most about what happens when the model is wrong.
How to apply
Email what you have built and why this role. No form, no cover letter format.
Apply by EmailContact lucas@isper.ai