Senior Voice AI Engineer - Real-Time Telephony & NLU

Mira Mace; San Francisco (CA); On-site; USD 140,000 - 200,000; Full time

Job summary

Mira Mace is seeking an engineer to own our voice AI end-to-end — the calling pipeline and the quality of every conversation. The role covers real-time telephony, streaming speech-to-text, LLM orchestration, and text-to-speech, with emphasis on latency budgets and robustness for aging audiences.

You will build evaluation harnesses, regression suites, and tooling to explain why a call went wrong, working closely with founders and daily with the voice team.

Qualifications

• Built and shipped real-time voice AI to production with actual calls.
• Hands-on depth across voice stack: streaming STT, TTS, endpointing, latency optimization.
• Experience running telephony in production (SIP/PSTN, Twilio or similar) and the failure modes.
• Built evaluation for voice/conversational systems.
• Debugged AI systems in production tracing audio, transcripts, and model behavior.
• Comfortable with ambiguity and building a playbook from scratch.

Responsibilities

• Own the voice pipeline end-to-end: telephony, streaming STT, TTS, endpointing, latency budgets.
• Drive call quality: define metrics, build evaluation harness, per-turn scoring.
• Handle real-world telephony: menus, holds, transfers, noisy lines, older audiences.
• Scale calling volume to thousands of calls/day with human handoffs when needed.
• Turn real conversations into prompts and pipeline changes quickly.
• Ship fast: production deployment and rapid iteration with real users.
• Lay foundation for scale: architecture robust to hundreds of calls/day to hundreds of thousands.

Skills

Voice AI

Real-time stack

Speech-to-text

Text-to-speech

LLM orchestration

Latency optimization

Production debugging

Tools

Twilio

SIP/PSTN

Job description

Mira Mace is seeking an engineer to own our voice AI end-to-end — the calling pipeline and the quality of every conversation. The role covers real-time telephony, streaming speech-to-text, LLM orchestration, and text-to-speech, with emphasis on latency budgets and robustness for aging audiences.

You will build evaluation harnesses, regression suites, and tooling to explain why a call went wrong, working closely with founders and daily with the voice team.