We’re building human-like, real-time voice models focused on natural turn-taking, interruption handling, and low-latency speech.

What you’ll do

• Design and generate conversational audio training data
• Train and fine-tune audio / speech models
• Build evaluation for latency, overlap, and interruption
• Own the loop: data → training → eval → production

What we’re looking for

• Strong ML background (PyTorch)
• Experience with audio or speech models
• Solid intuition for timing, latency, and real-time systems
• Startup / ownership mindset

Nice to have

• TTS, ASR, speech-to-speech, or streaming inference experience

Competitive comp + meaningful equity. Founding-level ownership.