Staff Conversational Designer (Fully Remote)

Deepgram; United States; Remote; USD 120,000 - 180,000; Full time

Job summary

Deepgram is seeking a senior designer/engineer to own conversational behavior for Deepgram voice agents, defining persona, turn-taking, repair, and pacing to feel natural.

You will collaborate with ML and Engineering on endpointing and latency budgets, build evals and transcript review practices, and establish developer design guidelines to scale conversational quality across teams. This role helps shape how customers interact with real-time APIs.

Qualifications

• Deep experience designing conversational behavior for LLM-based voice agents, not only scripted flows.
• Fluent with speech pipeline and a track record shipping turn-taking and repair patterns.
• Exceptional writing and ability to influence engineering partners without authority.

Responsibilities

• Define persona and voice system, design interruption and confirmation patterns, and own perceived responsiveness against latency budgets.
• Establish quality measurement through evals and structured failure analysis.
• Partner with ML and Research on ASR/TTS criteria and set design principles adopted across the team.

Skills

Conversational design

Voice agent persona design

Latency budgeting

Evaluation design

Job description

• Own conversational behavior for Deepgram voice agents, defining persona, turn-taking, repair, and pacing.
• Work directly with ML and Engineering on endpointing, barge-in, and latency budgets to make behaviors real.
• Build evals, transcript review practices, and developer design guidance to scale conversational quality.

What You’ll Do:

• Define persona and voice system, design interruption and confirmation patterns, and own perceived responsiveness against latency budgets.
• Establish quality measurement through evals and structured failure analysis.
• Partner with ML and Research on ASR/TTS criteria and set design principles adopted across the team.

Who You Are:

• Deep experience designing conversational behavior for LLM-based voice agents, not only scripted flows.
• Fluent with speech pipeline and a track record shipping turn-taking and repair patterns.
• Exceptional writing and ability to influence engineering partners without authority.

Deepgram

Deepgram is the leading platform for the Voice AI economy, providing real-time APIs for speech-to-text and text-to-speech, enabling organizations to build production-grade voice agents. Backed by a recent Series C and processed over 50,000 years of audio, Deepgram is a growth-stage company with an AI-first culture where innovation and rapid adaptation are key.